Issuu

Singapore Comparative Law Review Law Journal

United Kingdom Singapore Law Students Society (UKSLSS)

2026

Here to give you home Asia’s unified advantage winning team Lawyers who know Asia

At Rajah & Tann Asia, we are more than 1,000 strong across leading firms in ten countries, united by a shared commitment to excellence and regional insight. From headline-making cases to cross-border collaborations, you will gain exposure to the legal challenges shaping Asia today. We are lawyers who know Asia – are you ready to take the first step with a winning team? Find out more - Scan the QR code or visit https://www.rajahtannasia.com/joinus.

RAJAHTANNASIA.COM

DEDICATED. DRIVEN. DYNAMIC. With 700 legal professionals across Asia, you will be working with thought leaders with the ability to provide integrated and holistic legal advice. Renowned legal directories and publications consistently recognise us for our involvement in the region’s most significant transactions and high-profile disputes. We are A&G Asia.

42 ranked practices and 94 recognised lawyers across A&G Asia

Highest number of Tier 1 rankings across Singapore practice areas

41 ranked practices and 75 recognised lawyers across A&G Asia

Nine-time winner of the Singapore Corporate and Domestic Law Advisers Award since 2010

24-time winner of the National Firm of the Year (Singapore) award and most recently affirmed in 2025

Winner of the Regional Law Firm of the Year award (2021), Singapore Law Firm of the Year award (2023), and Innovative Law Firm of the Year award (2026)

Eight-time winner of the SE Asia Law Firm of the Year (most recently affirmed in 2026) and two-time winner of the Singapore Law Firm of the Year award

18 ranked practices and 45 recognised lawyers across A&G Asia Winner of the Asia Firm of the Year award (2026)

Winner of multiple practice area Firm of the Year awards

www.agasia.law

TABLE OF CONTENTS Front Matter Table Of Contents ........................................................................................................................... 1 President’s Address ......................................................................................................................... 3 Editor-In-Chief’s Address ............................................................................................................... 6 UKSLSS Executive Committee 2025/2026 .................................................................................... 8 UKSLSSEditorial Committee ......................................................................................................... 9 Patron’s Foreword ......................................................................................................................... 10 SCLR 2026 Contributions Converging Or Diverging Paths? The Governing Law Of Arbitration Agreements Under The UK Arbitration Act 2025 And Singapore’s Proposed IAA Reforms .................................................. 22 Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence: Inventive Step In Singapore, Uk And European Patent Law .................................................................................... 38 Chaos In Action: Reassessing Singapore’s Approach To Cryptoasset Classification And Defining The Boundaries Of The Tertium Quid .......................................................................................... 97 Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification — A Jurisprudential Analysis ............................................................................... 127 Statutory Sutures For The Liability Gap: A 'No-Fault' Prescription For Black-Box Ai In Singapore Healthcare ................................................................................................................................... 168 Unbroken Will – Assessing The Future Of Voluntariness In Singapore’s Criminal Procedure Code And Admissibility Regime ......................................................................................................... 192 Lawful Act Duress: A Comparative Study Of England And Singapore .................................... 213

In Search Of Meaning: The Economic Torts .............................................................................. 226 The Quistclose Trap: Intention In Quistclose Trusts .................................................................. 247 Recovering The Beneficiaries’ Profit From Their Fiduciary ...................................................... 271 Defending The Quasi-Employee Doctrine In Singapore In The Aftermath Of Bird V Dp ........ 293 “How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore .............. 317

Singapore Comparative Law Review

President’s Address

PRESIDENT’S ADDRESS Dear Readers, This year marks the 28th anniversary of the United Kingdom Singapore Law Students’ Society (“UKSLSS”) and the publication of the 21st Edition of the Singapore Comparative Law Review (“SCLR”). I would like to begin by thanking our Editor-in-Chief, Natasha Wong, and the Editorial Committee for the tremendous amount of work they have put into this year’s Review. Their dedication throughout the research, editing, and publication process has allowed the SCLR to continue providing students with a platform to engage critically with developments in Singapore law and beyond. A Recap of the 2025/2026 Academic Year Over the past year, the UKSLSS has continued to strengthen its role as a bridge between Singapore’s legal community and Singaporean law students studying in the United Kingdom. Through our events and initiatives, we have sought to provide our members with greater access to the legal profession while, at the same time, building a supportive community for Singaporean students studying abroad. We kicked off the year with our annual Legal Careers Helpdesk, through which students were able to receive guidance from those who had successfully navigated the vacation scheme, training contract and Bar application processes. I would like to extend my sincere thanks to our coordinators, Nicole Lim and John Yap, whose hard work was instrumental in bringing the event together. This year, we also launched our inaugural Legal Connect: Singapore 2026, bringing together law students with trainees and junior associates practising in Singapore. The event provided students with the opportunity to speak directly with practitioners in the early stages of their careers and gain a better understanding of the transition from university to legal practice in Singapore.

Singapore Comparative Law Review

President’s Address

Beyond our career initiatives, we continued to place great importance on fostering a sense of community amongst our members. Our Chinese New Year Dinner brought Singaporean students together to celebrate away from home, while our annual Freshers’ Tea welcomed incoming students before the beginning of their studies in the United Kingdom. These events remain an important part of the Society’s work and reflect our commitment to ensuring that the UKSLSS serves not only as a professional platform, but also as a community for our members. We have also continued to strengthen our relationships with the legal profession. This is reflected in the growth of our sponsorship portfolio, comprising two Gold sponsors and fourteen Silver sponsors. Their support has enabled us to expand the opportunities and initiatives available to our members. We supported multiple in-person recruitment events throughout the year, including the Rajah & Tann Meet & Mingle and WongPartnership Recruitment Tea, providing students with further opportunities to engage directly with practitioners and learn more about legal practice in Singapore. Our activities culminated in the Singapore Legal Forum 2026, our flagship event. This year’s Forum welcomed around 300 attendees, the highest number in the history of the UKSLSS, and brought together 17 law firms and partner organisations. We were also deeply honoured to welcome Justice Philip Jeyaretnam as our Guest of Honour. The scale of this year’s Forum was a fitting conclusion to the Society’s work over the past year and reflected the continued growth of both the event and the UKSLSS more broadly. The 21st Edition of the Review Turning to the 21st Edition of the Singapore Comparative Law Review, I would first like to express our sincere gratitude to our Patron, Justice Philip Jeyaretnam, for his support of this year’s Review. We are deeply honoured to have his guidance and contribution to the SCLR, and greatly appreciate the time he has taken to engage with the work of our student authors. This year’s Review reflects the wide range of questions confronting the law today. Across twelve articles, our contributors examine areas spanning international arbitration, intellectual property, criminal procedure, trusts, tort and consumer regulation, while also engaging with newer challenges arising from artificial intelligence and digital assets. The breadth of this year’s

Singapore Comparative Law Review

President’s Address

contributions demonstrates the continuing value of comparative legal analysis in considering not only how different jurisdictions approach similar problems, but also how established principles can respond to new technological and commercial developments. Natasha Wong, our Editor-in-Chief, shall provide a more detailed introduction to the contents of this year’s edition. I would like to recognise the tremendous amount of work that Natasha, the editorial team and all of our authors have invested in the Review. Their efforts have produced an edition which I hope will encourage readers to engage critically with both familiar areas of law and the new questions that continue to emerge around them. In closing, I would like to thank my Executive Committee and the wider UKSLSS committees. Every event, initiative and opportunity that the Society has delivered this year has depended on the time and effort that each of you has willingly given. It has been a privilege to work alongside you, and I am immensely grateful for everything that we have been able to accomplish together. I trust that this year’s Review will give you much to reflect on as you continue your legal studies. Beyond the Review, I hope that the UKSLSS has been a source of community and connection throughout your time in the United Kingdom and Singapore. As you move on to the next stage of your studies and careers, I wish you every success and all the very best for what lies ahead. Yours sincerely, Joshua Chang 26th President of the United Kingdom Singapore Law Students’ Society

Singapore Comparative Law Review

Editor-In-Chief’s Address

EDITOR-IN-CHIEF’S ADDRESS Dear Readers, The Singapore Comparative Law Review, the flagship publication of the United Kingdom Singapore Law Students' Society ("UKSLSS"), holds its value in two aspects above all. Firstly, it offers a platform for students to write for other students, sustaining a cycle of academic engagement that runs on its own community's effort. Secondly, in doing so, the Review captures a running record of what its authors and readers have found worth deliberating, discussing and writing about each year. With that in mind, I am honoured to present to you the 21st edition of the Review. Last year's foreword closed with the hope that future editions might see more authors turn to private law, and this year's cohort has answered that call. Where recent editions have leaned heavily toward constitutional and administrative law, this edition is anchored by a strong cluster of private law scholarship spanning contract, tort, trusts and fiduciary obligations, employment, and personal property. I am proud that the Review has been able to reflect this shift in interest among our authors, and I hope it signals a more even balance between public and private law scholarship in the editions to come. This edition also marks a momentous change in the Review's patronage. For many years, the Review has been fortunate to have former Chief Justice Chan Sek Keong as its Patron, and his dedication and generosity toward every edition he graced has shaped what this Review has become. On behalf of the Editorial Committee, I want to record our deepest thanks to him as he steps down from this role. We are honoured that Justice Philip Jeyaretnam has agreed to succeed him as Patron for this edition, bringing his own distinguished experience on the Bench and in international arbitration to bear on the Review's articles, and we are deeply grateful for his generosity of time and spirit in taking on this role. This edition would not exist without the sustained work of many people: our sponsors, whose support makes the Review's publication possible each year; the authors, who carried these pieces

Singapore Comparative Law Review

Editor-In-Chief’s Address

from pitch to polished article over many months; and the editors, managing editors, and Executive Committee of the UKSLSS, whose work is felt on every page even where it goes unseen. Whilst putting this Review together, I have learnt that studying the law comparatively does not simply teach you another jurisdiction, but often teaches you your own more clearly. I hope that as you read through these pages, you arrive at the same appreciation. Yours sincerely, Natasha Wong Editor-in-Chief of the 21st Singapore Comparative Law Review

Singapore Comparative Law Review

UKSLSS Executive Committee 2025/2026

UKSLSS EXECUTIVE COMMITTEE 2025/2026

Joshua Chang President

Min Sittman Vice-President

Annika Yeo Finance Director

Neil Karnik General Secretary

Christian Ng Sponsorships Director

Emily Leow Marketing Director

Natasha Wong Editor-In-Chief

Singapore Comparative Law Review

UKSLSS Editorial Committee

UKSLSS EDITORIAL COMMITTEE

Editor-In-Chief Natasha Wong Managing Editors Alysa Lee Mynn

Asher Dharsan

Sijia Cai

Sit Jie Ren Editors

Jia Li Lim

Joshua Wan

Joshua Wong

Julius Goh

Li Tian Qi

Sean Smart

Seraphine Lai

Sri Ranjana

Su Abdullah

Taryn Liew

Victor Sam

Zhong Rui Yang

Singapore Comparative Law Review

Patron’s Foreword

PATRON’S FOREWORD This is my first year of as patron of the Society, and it is with great pleasure that I write the foreword of the Singapore Comparative Law Review 2026. This year marks the Bicentennial of the Second Charter of Justice 1826, which established the Singapore legal system. Under it, English statutes of general application then existing as well as the common law of England, were received, subject however to local circumstances. Unsurprisingly, much has changed in two hundred years. Not only has the world undergone immense geopolitical changes but so has technology. Society, including business practice, are very different. This is reflected in the range of articles that appear in this year’s issue, and which I have had the pleasure of reading. Three articles relate to AI and one concerns digital assets. AI especially continues to advance exponentially and it is timely for lawyers to grapple with the legal issues it raises. Other articles examine enduring but equally important issues of law, and I was particularly pleased to read articles in the field of equity and trusts, articles that showed both scholarship and insight. I now set out brief notes for each of the articles. 1. Converging Or Diverging Paths? The Governing Law Of Arbitration Agreements Under The UK Arbitration Act 2025 And Singapore’s Proposed IAA Reforms Author: Asher Dharsan The article compares the sharply different default rules adopted or proposed in the United Kingdom and Singapore for determining the law governing an arbitration agreement when parties have not expressly chosen one. Section 6A of the UK Arbitration Act 1996, in force from August 2025, adopts a binary rule: an express choice governs, failing which the law of the arbitral seat applies. SIDRA’s proposed reform of Singapore’s International Arbitration Act instead creates a three-tier hierarchy: an express choice for the arbitration agreement; the expressly chosen law of the main contract; and, only in the absence of both, the law of the seat. The article situates this divergence against the earlier Sulamerica, Enka and BCY jurisprudence, and evaluates it through party autonomy, separability, commercial expectations and the New York Convention. The UK rule offers clarity, predictability and close alignment with the supervisory law of the seat and the Convention’s enforcement framework. It may also prevent a hostile main-contract

Singapore Comparative Law Review

Patron’s Foreword

law from invalidating the arbitration clause. Yet it risks treating the parties’ procedural choice of seat as a substantive choice of law and may simply relocate litigation to the question of what constitutes an “express agreement”. Singapore’s proposed main-contract-law default better reflects the ordinary commercial expectation that one legal system governs the contractual relationship as a whole and avoids an artificial fragmentation between the main contract and its arbitration clause. Its weakness is rigidity: converting the present defeasible inference into a hard statutory default could defeat an otherwise valid arbitration agreement. The article ultimately prefers Singapore’s approach, but recommends that the main-contractlaw default be rebuttable so that courts can preserve the validation principle and account for contrary contextual indications. Its broader conclusion is practical rather than doctrinal: neither statutory regime can eliminate uncertainty caused by incomplete drafting. Parties, institutions and practitioners should therefore specify expressly the law governing the arbitration agreement in model clauses and transaction documents. 2. Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence: Inventive Step In Singapore, UK And European Patent Law (Runner Up for Best Article Award 2026) Author: Joia Ng Yi Jing The article examines how routine use of artificial intelligence in research challenges patent law’s traditional benchmark for inventive step: the notional person skilled in the art. It identifies two opposing risks. If the skilled person is imagined without AI tools already used by ordinary researchers, routine AI-assisted outputs may wrongly appear inventive. If the benchmark is instead endowed with frontier AI capabilities, genuinely difficult advances may be treated as obvious. Drawing on Singapore and UK doctrine and EPO practice, the article argues that the benchmark should remain a human practitioner or multidisciplinary team, while its attributed knowledge, resources and working methods evolve with ordinary technical practice at the relevant priority date. The proposed “AI-augmented skilled person” rests on three evidential considerations. A particular AI tool should affect the inventive-step baseline only if ordinary practitioners could realistically access it, routinely used and accepted it as reliable for the relevant task, and could deploy it without exceptional expertise, data or infrastructure. The article distinguishes common general knowledge—what practitioners know and accept—from the wider technical means available to them. Even when an AI-assisted route forms part of ordinary practice, that does not make every result obtained through it obvious: courts and patent offices must still 11

Singapore Comparative Law Review

Patron’s Foreword

assess problem framing, technical choices, expected success, alternative routes, experimental burden and unexpected effects. The comparative analysis concludes that existing law is flexible enough to accommodate this approach without statutory reform. Singapore is especially well placed to develop it because section 15 of the Patents Act draws on UK jurisprudence, IPOS already uses supplementary AI guidance, and EPO materials provide a useful distinction between common knowledge and the means for routine work. The article therefore recommends incremental clarification through IPOS guidance and judicial reasoning, supported by contemporaneous, field-specific evidence. This would preserve a human-centred benchmark while preventing both over-patenting of routine AI outputs and an unrealistically high threshold based on exceptional machine capability. 3. Chaos In Action: Reassessing Singapore's Approach To Classification And Defining The Boundaries Of The Tertium Quid (Honourable Mention for Best Article Award 2026) Authors: Davon Pung, Shanice Yee

Cryptoasset

This article argues that Singapore’s recognition of cryptoassets as property has outpaced the development of a coherent framework for classifying them. Singapore courts have protected cryptocurrencies and NFTs through proprietary remedies, but their decisions have been piecemeal and often interlocutory. In particular, ByBit treated USDT as a chose in action, relying on a broad and flexible conception of that category. The authors contend that this approach is doctrinally unsatisfactory: decentralised cryptoassets such as Bitcoin and Ether are neither things in possession, because they are intangible and not physically possessable, nor choses in action in the orthodox sense, because they do not embody an enforceable claim against an identifiable counterparty. Stretching either category to accommodate them would distort established legal doctrines concerning possession, conversion, transfer and remedies. Comparing Singapore with the United Kingdom, the article notes that the UK’s Property (Digital Assets etc) Act 2025 deliberately confirms that an asset may be personal property even if it falls outside the traditional binary, while leaving the detailed incidents of such property to the common law. The authors regard this as preferable to Singapore’s incidental judicial development but argue that the UK legislation is too minimalist because it supplies no criteria for identifying third-category assets. That lack of guidance risks uncertainty over remedies, tracing, insolvency and the persistence of an asset’s identity through transfers, as illustrated by early post-Act litigation.

Singapore Comparative Law Review

Patron’s Foreword

The article therefore proposes that Singapore enact legislation or statutory guidance recognising a third category of personal property and adopting a two-layer classificatory framework. The first layer comprises four core indicia: rivalrousness; existence independently of the legal system; control through cryptographic mechanisms; and the absence of an identifiable counterparty obligor. For borderline cases, courts would additionally consider susceptibility to unauthorised factual interference and persistence of identity through transfer. This framework would ordinarily place Bitcoin, Ether and NFTs in the third category, while leaving issuer-backed assets such as tokenised debt securities within choses in action and requiring closer analysis of stablecoins and other hybrid assets. The authors conclude that this approach would give courts and market participants greater clarity while preserving the common law’s flexibility to refine the legal incidents of digital assets as technology evolves. 4. Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification — A Jurisprudential Analysis Author: Michelle Yap This article examines whether AI chatbots should be legally permitted to generate sexual content for adults' private gratification, a question the author argues is increasingly urgent given that technological developments are outpacing regulatory responses. After defining key terms and scoping the analysis to six categories of adult sexual content — ranging from nonconsensual deepfakes to fully synthetic content, with and without intent to disseminate — the article surveys the regulatory positions of Australia, China, Singapore and the United Kingdom. It finds that China alone imposes an outright prohibition, while the other three jurisdictions largely permit such content for private adult use, albeit with varying degrees of restriction, primarily through age assurance requirements rather than substantive prohibitions. The article then applies four jurisprudential frameworks to the question. Mill's harm principle and Hart's modified liberalism both tend towards permissibility, though the author identifies significant weaknesses in both, particularly their difficulty in accounting for indirect moral harm and the arbitrariness of the self-regarding/other-regarding distinction. Rawls's political liberalism similarly tends towards permissibility but is criticised for falsely claiming neutrality on conceptions of the good. By contrast, Finnis's natural law theory and virtue ethics — the frameworks the author ultimately endorses — support prohibition, on the grounds that such content undermines basic human goods, erodes human dignity, fosters disordered desire, and detracts from the common good.

Singapore Comparative Law Review

Patron’s Foreword

The author concludes that AI chatbots should not be legally permitted to generate sexual content for adults' private gratification, supported by practical considerations including the ineffectiveness of age assurance mechanisms, the disproportionate harm to women and children, and the risk of AI developers profiteering from vice. Importantly, the author does not advocate a one-size-fits-all response such as universal criminalisation, recognising the principle of subsidiarity and the value of non-regulatory tools such as education and ethical guidelines. The article closes by calling for shared responsibility across developers, regulators, institutions and citizens in ensuring that AI serves human dignity rather than undermining it. 5. Statutory Sutures For The Liability Gap: A 'No-Fault' Prescription For BlackBox AI In Singapore Healthcare Author: Jolyn Ho Min Yee The article argues that Singapore’s negligence framework is strained when medical injury arises from opaque “black-box” AI whose reasoning cannot be meaningfully examined by clinicians, patients, or courts. Although the Bolam-Bolitho standard can still regulate clinicians who blindly rely on AI without corroborative checks, doctors should not become “liability sponges” where they have exercised reasonable human oversight but cannot inspect the model’s internal logic. Hospitals and developers may owe duties to validate, monitor, and train these systems properly, yet patients face formidable obstacles in obtaining proprietary evidence and proving which defect or actor caused the harm. Using New Zealand’s Accident Compensation Act as a functional comparison, the article finds that a no-fault scheme offers faster and more accessible compensation because claimants need not prove negligence and may establish causation through robust inference or material contribution. However, the New Zealand model has weaknesses: exclusions for injuries substantially caused by an underlying condition can leave high-risk patients uncompensated, fault reasoning may re-enter through the “ordinary consequence” inquiry, and developers face little meaningful deterrence because exemplary damages require proof of subjective recklessness. The proposed solution is a Singapore-specific, sectoral no-fault fund with a triage mechanism. Negligence would continue to apply where a clinician had a meaningful opportunity to intervene, while claims would enter the fund where the clinician properly discharged that role and the failure lay beyond meaningful human scrutiny. Developers and deployers would finance the fund through risk-adjusted contributions tied to compliance with AI governance standards; compensation would be apportioned rather than wholly barred for patients with 14

Singapore Comparative Law Review

Patron’s Foreword

serious underlying conditions; avoidability would be assessed without reintroducing fault; and a sunset clause would permit a return to ordinary tort principles once explainable AI becomes sufficiently reliable and auditable. 6. Unbroken Will – Assessing The Future Of Voluntariness In Singapore’s Criminal Procedure Code And Admissibility Regime Author: Nickolaus Ng The article defends Singapore’s confession-admissibility regime under section 258 of the Criminal Procedure Code and rejects replacing its voluntariness test with the United Kingdom’s reliability-centred approach under the Police and Criminal Evidence Act 1984. It explains that Singapore’s framework rests on three overlapping rationales: excluding statements made unreliable by coercion or inducement, protecting accused persons from improper pressure, and discouraging investigative misconduct. Although the voluntariness test is technical and does not capture every potentially unreliable confession, its combined subjective and objective elements already allow courts to assess whether threats, promises, inducements, or oppression realistically undermined the accused’s freedom of choice. The article contrasts this framework with the United Kingdom’s move from common-law voluntariness to PACE, under which a confession is excluded if police conduct was likely to render it unreliable, with oppression remaining an independent ground. It contends that the historical reasons for reform in England do not translate well to Singapore: the old English rule could exclude confessions upon very slight inducements, whereas Singapore’s test already demands an objective assessment of the seriousness and effect of the alleged inducement. In practical terms, both PACE and Singapore law require a fact-sensitive inquiry into police conduct and its likely influence, so importing PACE would produce little substantive benefit while risking confusion between admissibility and the ultimate truth or weight of the confession. Singapore also retains important safeguards that make wholesale reform unnecessary. The Prosecution must prove voluntariness beyond reasonable doubt once it is challenged, and courts may exclude technically admissible statements where procedural irregularities or other circumstances make their prejudicial effect outweigh their probative value. Given that a conviction may rest on an uncorroborated confession and that some offences carry capital punishment, the article concludes that preserving the structured voluntariness test is safer than adopting an indeterminate reliability standard. The existing regime, supplemented by judicial

Singapore Comparative Law Review

Patron’s Foreword

discretion, adequately balances evidential integrity, protection of accused persons, discipline of law enforcement, and investigative efficiency. 7. Lawful Act Duress: A Comparative Study Of England And Singapore Author: Chuan Wee Foo This article examines whether lawful conduct can amount to duress and compares the boundaries of the doctrine in England and Singapore. Both jurisdictions recognise that lawful acts may exceptionally constitute illegitimate pressure, but they differ markedly in approach. In Times Travel, the UK Supreme Court adopted a narrow position, largely confining lawful act duress to threats involving the exploitation of criminal wrongdoing or vulnerability deliberately created through illegitimate means to procure the waiver of a civil claim. Singapore’s approach, illustrated by Oon Swee Gek and informed by the Enonchong factors, is broader and permits an assessment of whether a lawful threat was made in bad faith, was unreasonable, abused legal process, or was unconscionable in all the circumstances. The article argues that lawful act duress is normatively justified. It rejects the objections that the doctrine unduly infringes freedom of contract or is too uncertain, reasoning that contractual freedom is not absolute and that lawful conduct may still be used illegitimately to procure defective consent. Nor should such cases merely be absorbed into undue influence or unconscionability, since doing so would expand those doctrines and obscure the distinctive wrong addressed by lawful act duress. Ultimately, the article prefers the Singaporean model because it better balances contractual certainty with substantive fairness. The English test is criticised for placing excessive weight on enforceability before asking whether consent was validly obtained. Singapore’s more contextual inquiry is said to respond better to attempted exploitation while preserving certainty through a high threshold: lawful act duress remains confined to very rare cases. 8. In Search Of Meaning: The Economic Torts (Honourable Mention for Best Article Award 2026) Authors: Kai En Ho and Parth Anand This article asks whether the economic torts share a single coherent rationale. It distinguishes thin functional coherence—the common concern with deliberate interference in economic interests and the permissible limits of competitive pressure—from doctrinal or justificatory unity. Although inducing breach of contract, causing loss by unlawful means, intimidation, and the conspiracy torts may sensibly be studied together, they protect different interests and

Singapore Comparative Law Review

Patron’s Foreword

employ different structures. Inducing breach is accessorial and depends on a contractual breach; the unlawful means tort imposes primary liability for targeted wrongdoing against a third party; and conspiracy turns on combination, unlawful means, or a predominant purpose to injure. The article contends that attempts at unification, particularly the Lindley–Hoffmann rationale developed through Quinn v Leathem and OBG v Allan, cannot explain the whole field. Persistent fractures remain over the meaning of “unlawful means”, the required intention, the distinction between two-party and three-party liability, and the relationship between primary and accessory liability. These differences are sharpened by the interaction of the torts with competition law, labour law, contract, and company law. A general tort of intentional economic harm would therefore amount to major reform, not an accurate statement of existing doctrine, and could undermine predictability or bypass choices made in adjacent statutory regimes. For Singapore, the article recommends “disciplined differentiation” rather than importing a supposed unity from English law or adopting a broad prima facie liability rule. Courts should identify, tort by tort, the interest protected, the form of wrongful interference, the relevant mental element, and the surrounding legal regime. On this account, the economic torts have enough functional coherence to be coordinated, but their defensible coherence lies in transparent, principled distinctions rather than a single master doctrine. 9.

The Quistclose Trap: Intention In Quistclose Trusts (Winner of Best Article Award 2026) Author: Navin Sivakumar

This article argues that the most coherent test for a Quistclose trust is Lord Millett’s formulation in Twinsectra v Yardley: whether the parties intended the transferred assets to be at the recipient’s “free disposal”. It traces the doctrine from Barclays Bank v Quistclose Investments, criticising Lord Wilberforce’s account of successive primary and secondary trusts, and shows how Twinsectra recast the arrangement as one in which the transferor retains the beneficial interest while the recipient has only a limited power to apply the fund for the agreed purpose. The article rejects later tests requiring an intention to create a trust, an intention to retain a beneficial interest, or the recipient’s subjective “proprietary understanding”. Those approaches either collapse the Quistclose trust into an express trust, beg the very question of beneficial ownership, or undermine objective certainty. Instead, the author grounds the free-disposal test in unconscionability: equity intervenes because a recipient who accepts a specific fund on restricted terms should not be permitted to treat it as part of their general assets.

Singapore Comparative Law Review

Patron’s Foreword

The practical inquiry should therefore focus on whether the recipient had free disposal of the particular assets. A stated purpose is necessary but insufficient; relevant indicators include segregation from general funds, the specificity and exclusivity of the purpose, and the wider commercial relationship. This approach, the author concludes, gives English and Singapore law a principled basis for recognising Quistclose trusts without becoming trapped in rigid taxonomic debates. 10. Recovering The Beneficiaries’ Profit From Their Fiduciary Author: Meng Lingyi This essay examines whether a fiduciary should be required to disgorge profits only where the breach was a but-for cause of those profits. It contrasts the Singapore Court of Appeal’s decision in UVJ v UVH, which adopted a but-for causation requirement, with the UK Supreme Court majority’s approach in Recovery Partners GP Ltd v Rukhadze, which asks instead whether the profit is significantly attributable to the fiduciary position, its advantages, or conflicted activity. The author contends that UVJ wrongly imports the methodology of equitable compensation into the distinct, gain-based remedy of an account of profits. Compensation is loss-focused and naturally requires counterfactual causation; disgorgement is prophylactic, aimed at preserving single-minded loyalty and removing temptation. A but-for defence allows fiduciaries to argue that the same gain could have been earned lawfully, thereby weakening the strict no-profit rule and reopening space for self-serving rationalisation. The essay recommends that Singapore develop an attributional causation standard rather than simply transplant the divided reasoning in Rukhadze. The principal should first show that the profit arose by reason of the fiduciary position; the burden should then shift to the fiduciary to disprove the connection or justify an equitable allowance for legitimate skill, labour, or expenditure. This, it argues, would protect institutional trust, address evidential asymmetry, and preserve fairness without diluting the deterrent force of fiduciary law. 11. Defending The Quasi-Employee Doctrine In Singapore In The Aftermath Of Bird v DP Authors: Lam Yu Han and Sim Suan Yeong, Amelita This article argues that Singapore should affirm the quasi-employee doctrine despite the High Court of Australia’s rejection of it in Bird v DP. The doctrine permits vicarious liability where a tortfeasor is not technically an employee but occupies a relationship sufficiently akin to

Singapore Comparative Law Review

Patron’s Foreword

employment. The authors contrast Bird’s insistence on “strict” employment with the UK approach in Various Claimants v Catholic Child Welfare Society and with Singapore decisions that have approved or applied the doctrine, although its status here is not yet conclusively settled. The authors submit that the objections underlying Bird do not translate to Singapore. Singapore accepts compensation, enterprise risk, and deterrence as foundations of vicarious liability; its courts remain institutionally competent to develop the common law because Parliament has not occupied the field; and its multifactorial employment test already looks beyond contractual labels to control, integration, and economic reality. In that setting, an undefined requirement of “strict” employment is said to be formalistic, arbitrary, and potentially circular. The doctrine is also defended as necessary to keep vicarious liability effective amid modern organisational forms, including unincorporated associations, platform work, the gig economy, and potentially artificial intelligence. Because it remains subject to a fact-sensitive qualifyingrelationship inquiry and the separate requirement that the tort be closely connected with the work, it can provide compensation, distribute enterprise risk, and promote deterrence without imposing unlimited liability. The article therefore urges Singapore courts not to follow Bird. 12. “How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore Author: Edison Toh This article evaluates whether Singapore’s regulation of multi-level marketing meaningfully distinguishes legitimate direct selling from recruitment-driven pyramid schemes. Although the Prohibition Act broadly bans pyramid selling and the Exclusion Order permits schemes satisfying prescribed safeguards, the author argues that the framework concentrates too heavily on legal form—especially whether rewards are formally attributed to product sales—rather than on the economic source of sales and participant earnings. Modern product-based pyramid schemes may comply formally while remaining dependent on continual recruitment. Qualification purchases, internal consumption, inventory loading, and downline volume can convert recruits into both sellers and captive purchasers, allowing product sales to perform the economic role once played by entry fees. The article argues that decisions such as Chua Hock Soon James v PP address straightforward recruitment schemes but do not resolve this harder boundary problem. It also questions the effectiveness of

Singapore Comparative Law Review

Patron’s Foreword

safeguards concerning records, misleading representations, buybacks, and complaint statistics, given persistent participant losses and enforcement difficulties. Comparing Singapore with the United States, the author favours supplementing the Exclusion Order with an “economic reality” inquiry into genuine retail demand, participant purchasing incentives, and the scheme’s viability without continuing network expansion. The article also proposes standardised earnings disclosures and licensing or competency requirements for recruiters. It ultimately questions whether MLMs should retain a categorical statutory exclusion at all unless the industry can demonstrate that its social and economic benefits justify the risks and that its revenues arise from genuine consumer demand. This issue once again demonstrates how useful and interesting a comparative law approach is. I commend it to the reader. Justice Philip Jeyaretnam Patron of the 21st Singapore Comparative Law Review

Lawyers who know Asia

WE ARE AS YOU ARE ambitious motivated passionate trustworthy well-rounded At Rajah & Tann Singapore, we have a bold vision of pioneering legal practice. We’re looking for someone who’s ready to shape the future of law together. If you share the same passion, get to know us better. Scan the QR code or visit our website: https://sg.rajahtannasia.com/careers-singapore for more details.

Singapore Comparative Law Review

Converging or Diverging Paths?

CONVERGING OR DIVERGING PATHS? THE GOVERNING LAW OF ARBITRATION AGREEMENTS UNDER THE UK ARBITRATION ACT 2025 AND SINGAPORE’S PROPOSED IAA REFORMS Asher Dharsan* ABSTRACT The law governing an arbitration agreement remains one of the most contested questions in international commercial arbitration, particularly where parties have failed to designate it expressly. Recent reforms in the UK and Singapore have brought this issue into sharper focus by adopting/proposing fundamentally different default rules. Section 6A of the UK’s Arbitration Act 1996 establishes the law of the arbitral seat as the statutory default, while the Singapore International Dispute Resolution Academy (SIDRA) has recommended a three-tier framework that gives priority to the governing law of the main contract before resorting to the law of the seat. This article undertakes a comparative doctrinal analysis of these competing approaches, examining their consistency with principles of party autonomy, separability, the New York Convention, and the commercial expectations of international contracting parties. It argues that Singapore’s proposed framework provides the more commercially realistic default as it better reflects the ordinary expectation that a single legal system should govern the contractual relationship in its entirety, including the arbitration agreement. However, the article contends that the proposed statutory presumption should be rebuttable rather than irrebuttable to preserve the validation principle and avoid undermining otherwise effective arbitration agreements. Although the English model offers greater certainty and closer alignment with the Convention’s enforcement architecture, neither regime can eliminate uncertainty caused by incomplete drafting. The article concludes that the enduring solution lies not in refining default conflict-of-laws rules, but in embedding an express designation of the governing law of arbitration agreements within standard commercial drafting practice.

King’s College London, LLB Class of 2026. I am grateful to the reviews from the Editorial Team for their guidance and comments on the earlier drafts of my article. All errors that remain are my own. Email for correspondence: asher.dharsan1@gmail.com

Singapore Comparative Law Review

Converging or Diverging Paths?

INTRODUCTION

The question of which law governs an arbitration agreement is among the most consequential and contested issues in international arbitration. It determines the procedural rules that apply to any dispute, which can shape the outcome of the dispute. Despite its significance, commercial parties rarely indicate the governing law of the arbitration agreement in contractual arrangements. 1 When contracts are silent, courts and tribunals must resolve it through default rules. Yet, guidance on the ‘default’ governing law is not uniform across jurisdictions. Some jurisdictions default to the law of the seat, while others default to the governing law of the main contract. Two recent reforms in the United Kingdom (“UK”) and Singapore have made this divergence even sharper. In force since 1st August 2025, Section 6A of the UK’s Arbitration Act (“AA”) 1996 establishes the law of the seat as the statutory default governing law of an arbitration agreement where parties have not made express provision for it. In contrast, the Singapore International Dispute Resolution Academy (“SIDRA”) recommended in November 2024 that Singapore’s International Arbitration Act (“IAA”) 1994 be amended to adopt a three-tier hierarchy: -

Tier 1: Any express law designated by the parties for the arbitration agreement itself.

Tier 2: The express governing law of the main contract containing the arbitration agreement.

Tier 3: Only in the absence of both, the law of the seat. 2

Both regimes agree on Tier 1 and retain the seat as an ultimate fallback. The divergence lies in Tier 2, as Singapore interposes the main contract’s chosen law as an intermediate default before the seat while the UK does not. As of April 2026, the SIDRA recommendation has not been enacted. The Ministry of Law conducted a public consultation from 21 March to 2 May 2025

Enka Insaat Ve Sanayi AS v OOO Insurance Company Chubb [2020] UKSC 38 [43]. Singapore International Dispute Resolution Academy (SIDRA), Review of the Singapore International Arbitration Act (2024) Issue 6, 99–119. 2

Singapore Comparative Law Review

Converging or Diverging Paths?

presenting three options: retain the common law position, adopt the SIDRA recommendation, or adopt the English seat-law approach.3 A bill has not been introduced. Should the SIDRA recommendations be adopted in full, the divergence in default laws between London and Singapore (the two most preferred seats for arbitration) may create uncertainty for commercial parties.4 A dispute arising from a contract governed by English law, which calls for a SIAC-seated arbitration, might now face the difficulty when determining the law governing the arbitration: is it Singapore law under Section 6A of the AA 2025, or English law under the SIDRAamended IAA 1994? This article examines the doctrinal basis and policy rationale of both regimes and evaluates which better serves international commercial parties. We argue that Singapore’s proposed SIDRA framework is the more commercially realistic approach, given its graduated three-tier hierarchy, as it better reflects the reasonable expectations of parties who intend a single system of law to govern their contractual relationship in its entirety. While Section 6A achieves formal certainty and aligns with the New York Convention’s enforcement architecture, it does so at the cost of imposing the law of the seat on an arbitration agreement whose parties had no such intention. We further argue that Singapore’s final enacted legislation should incorporate a rebuttable (rather than an irrebuttable) Tier 2 default, preserving the protective function of the validation principle that the current SIDRA formulation inadvertently eliminates. However, both reforms ultimately converge on one point of agreement. No statutory default can substitute for express party designation of the governing law of the arbitration agreement, and the persistence of this gap in commercial drafting practice is the structural problem that thought leadership in this field must now address. The comparison has particular significance for Singapore as international arbitration is a strategic pillar of Singapore’s positioning as a global commercial hub, 5 underpinning the jurisdiction’s attractiveness to multinational corporations, financial institutions, and state-linked enterprises

Singapore Ministry of Law, Consultation Paper on the Proposed Amendments to the International Arbitration Act 1994 (2025). ` 4 School of International Arbitration (Queen Mary University of London) and White & Case, ‘International Arbitration Survey: The path forward’ (2025), 7. 5 Queen Mary University of London and White & Case, International Arbitration Survey: The Path Forward — Realities and Opportunities in Arbitration (2025).

Singapore Comparative Law Review

Converging or Diverging Paths?

choosing a neutral seat in Asia. Adopting a clear default governing law rule which is closest to parties’ intention promotes commercial certainty, maintaining Singapore's position as a hub for arbitration in Asia. II.

DOCTRINAL BACKGROUND

It is well established that the overarching principles of separability and party autonomy govern an arbitration agreement.6 Separability of an arbitration agreement allows for the law governing the contract as a whole to differ from the arbitration agreement, while party autonomy implies that the choice of the contracting parties, implied or express, should be respected. In the absence of express provision, judges and arbitrators exercise their discretion to determine the parties’ intention in line with these principles. The contention arises where parties do not expressly state the law governing the arbitration contract, and the law of the seat differs from the law of the main contract. In such cases, the New York Convention and Model Law both suggest that the law of the country that the award was made (which will be the law of the seat of the arbitration) will preside over the arbitration clause.7 A. UK Before the AA 2025, the UK, however, took a markedly different approach. In Sulamerica,8 the courts formed a three-stage enquiry to determine the law governing the arbitration agreement: 1. First Limb: The express choice of parties’, 2. Second Limb: the implied choice of parties, and 3. Third Limb: the law with which the arbitration agreement has its ‘closest and most real connection’.

Alan Redfern, Martin Hunter et al, Redfern and Hunter on International Arbitration (7th edn, Oxford University Press 2023), para 3.36. 7 Convention on the Recognition and Enforcement of Foreign Arbitral Awards (adopted 10 June 1958, entered into force 7 June 1959) 330 UNTS 38, Art V(1)(a) and UNCITRAL Model Law on International Commercial Arbitration (United Nations, 1985, as amended 2006) UN Doc A/40/17, Art 16(1). 8 Sulamérica Cia Nacional de Seguros SA v Enesa Engenharia SA [2012] EWCA Civ 638 [25] (Moore-Bick LJ).

Singapore Comparative Law Review

Converging or Diverging Paths?

In that case, the parties had not indicated the law of the arbitration. While the companies were Brazilian, the courts found that applying Brazilian law would significantly undermine the arbitration agreement as it would require the consent of one of the parties. As there was no indication that the parties intended to enter into a one-sided arrangement, it was held that the arbitration agreement had the closest and most real connection with English law. This approach was affirmed and refined in Enka, where the Supreme Court determined the relative weight to be attached to the law of the seat of the arbitration where the governing law of the main contract and seats differ. Regarding the second limb, the Supreme Court extended the presumption to state that where there is a choice of law for the main contract, the parties impliedly chose the main contract law to govern the arbitration agreement as well. 9 However, they clarified that there was no presumption that by choosing a seat of arbitration the parties had chosen the law of the seat to govern the contract by implication. 10 The Supreme Court also clarified the application of the final limb of the enquiry, noting that in general, the arbitration agreement will be most closely connected with the law of the seat of the arbitration, with the sole exception being the case where the arbitration agreement would be invalid under the law of the seat but not the law governing the rest of the contract (such as in the case of Sulamerica above). Notably, in limiting the third limb, the Supreme Court clarified the English approach as a nuanced exception to the general approach consistent with the New York Convention in order to better account for the commercial intentions of parties. B. SINGAPORE The presumption in Sulamerica that the governing law of a contract applies also to its arbitration agreement was affirmed in BCY v BCZ,11 where the High Court justified their adoption of the approach based on commercial practicalities (to address this point in later recommendations): business parties rarely specify the law applicable to the arbitration agreement as distinct from the main contract. Concepts of separability, while theoretically applicable when drafting the arbitration agreement, were unlikely to be considered by business parties focused on the main contract itself. Put another way, at the time of signing a business agreement, commercial parties 9

Enka (n 1) [43]. Enka (n 1) [66], [94]. 11 [2016] SGHC 249 [59]. 10

Singapore Comparative Law Review

Converging or Diverging Paths?

rarely consider the specifics of ‘midnight clause[s]’ like the governing law of potential arbitrations which might never materialise. 12 The BCY framework was affirmed by the Court of Appeal in BNA v BNB13 and was more recently applied in a pre-award context.14 III.

THE RECENT REFORMS

A. UK Section 6A — Seat Law as Statutory Default Section 6A of the AA 2025 abolishes the Sulamerica Stage 2 implied-choice enquiry. The new provision is deliberately sparse: (1) The law applicable to an arbitration agreement is — (a) the law that the parties expressly agree applies to the arbitration agreement, or (b) where no such agreement is made, the law of the seat of the arbitration in question. (2) For the purposes of subsection (1), agreement between the parties that a particular law applies to an agreement of which the arbitration agreement forms a part does not constitute express agreement that that law also applies to the arbitration agreement.

The structure is binary: Section 6A(1)(a) governs express agreements, while everything else defaults to the seat under Section 6A(1)(b). Section 6A(2) performs a critical limiting function, preventing the choice-of-law clause in the main contract from constituting ‘express agreement’ to the arbitration agreement by virtue of that clause alone. This directly reverses the Enka majority’s holding that a governing law creates a rebuttable presumption at Tier 2 in favour of the main contract’s law.15 This reform also aligns the English legislative approach with Article 16.4 of the LCIA Rules on the law of the seat being the default. The UK Law Commission’s rationale for the reform rested on five pillars: ● Simplicity and predictability.

Redfern and Hunter (n 6), para. 2.04. [2019] SGCA 84 [45]–[47], [62]. 14 Anupam Mittal v Westbridge Ventures II Investment Holdings [2023] SGCA 1. 15 Enka (n 1) [143]. 13

Singapore Comparative Law Review

Converging or Diverging Paths?

● Reduction of satellite litigation ● Alignment with the curial law of the supervisory court. ● Consistency with France, Germany and Sweden, which already apply the seat-law default. ● Compatibility with Article V(1)(a) of the New York Convention 1958, which defaults to the law of the country where the award was made at the enforcement stage. 16 Section 6A(3)-(4) carves out investor-state arbitration agreements derived from standing offers in treaties or non-UK legislation, preserving the pre-existing conflict-of-laws framework for that distinct context. Section 6A(2) provides that a main-contract governing law clause ‘does not constitute’ express agreement for the arbitration agreement. Notably, the provision omits any ‘of itself’ qualifier, raising the question of whether a broadly worded clause of the Kabab-Ji17 or UniCredit18 type could nonetheless satisfy S6A(1)(a) when read with surrounding structural context. Pending litigation on this matter, there have been concerns that satellite litigation may migrate from ‘what is an implied choice?’ to ‘what is an express choice?’, which would ultimately undermine the certainty that Section 6A promises. A. Singapore’s SIDRA Proposal — Main Contract Law as First Default SIDRA published its Review of the Singapore International Arbitration Act (2024) in November 2024, with issue 6 proposing a three-tier statutory framework for the IAA. The normative shift underlying Tier 2 is more than a restatement of existing Singapore common law. Under BCY, Tier 2 of the Sulmerica test operated by interference. Courts reasoned that parties who had expressly chosen a governing law for the main contract must have intended that same law to govern the arbitration agreement contained within it. That inference was defeasible, as it could

UK Law Commission, Review of the Arbitration Act 1996: Final Report and Bill (Law Com No 413, HC 1787, 2023) ch 12, paras 12.68–12.75 (rationale), para 12.78 (recommendations). 17 Kabab-Ji SAL (Lebanon) v Kout Food Group (Kuwait) [2021] UKSC 48, [35], [39]–[44], [49]–[52]. 18 UniCredit Bank GmbH v RusChemAlliance LLC [2024] UKSC 30, [22], [32]–[60].

Singapore Comparative Law Review

Converging or Diverging Paths?

be displaced by the validation principle demonstrated in Anupam Mittal,19 or by other contextual factors pointing against it. SIDRA’s proposal structurally converts that defeasible implied choice inference into a statutory presumption provision triggered by the mere presence of an express governing law clause in the main contract without any further judicial enquiry into the intention of parties. The argument for this move is that it eliminates the unpredictability of Tier 2 of the enquiry while preserving its most common substantive outcome. However, the possible critique is that it may be over-inclusive. The statutory default will apply even in cases where a court conducting the BCY analysis would not have found a sufficient implied choice. For example, where the main contract’s governing law was chosen for purely regulatory reasons unrelated to the arbitration agreement, or where the seat was selected because its laws were intended to govern the arbitration clause. SIDRA essentially trades one form of uncertainty (implied-choice analysis) for a different form of rigidity (a presumption that may not reflect actual party intent in all cases). The broader justification for SIDRA's Tier 2 default is fourfold. On international outlook, Singapore-seated cases typically involve parties and contracts with predominantly foreign elements. Defaulting to Singapore law as a seat would impose a locally sourced governing law on parties who chose Singapore for its procedural infrastructure. On the choice of seat versus choice of substantive law, the selection of a seat reflects confidence in the supervisory court and institutional rules, not a preference for the seat’s substantive contract law. 20 On contractual coherence, applying different laws to the main contract and the arbitration agreement creates what Redfern and Hunter describe as ‘an unreal distinction’, fragmenting the legal framework governing a single transaction. Finally, in fidelity to party expectations, most commercial parties negotiating an international contract with a governing law clause expect that law to govern all clauses, including the agreement to arbitrate. 21

Anupam (n 14), [65]–[66], [70], [74]–[75]. Skrine, "Arbitration Law Reform in Singapore and the UK: A Comparative Overview for Malaysian Practitioners" (August 2025) <https://www.skrine.com/insights/alerts/august-2025/arbitration-law-reform-in-singapore-and-theuk-a-c> accessed 15 April 2026. 21 Redfern and Hunter (n 6), para 3.36. 20

Singapore Comparative Law Review

IV.

Converging or Diverging Paths?

NORMATIVE EVALUATION: WHICH REGIME BETTER SERVES COMMERCIAL PARTIES? A. Arguments for the UK Seat-Law Default

Four arguments support the seat-law default as the superior rule for international commercial arbitration. The first is certainty. The seat of arbitration is almost invariably fixed by the arbitration clause itself, making it an identifiable and uncontested anchor for the governing law enquiry. By contrast, the main contract’s governing law may itself be disputed, and resolving that prior question is neither straightforward nor cost-free. Legal certainty in this context carries measurable economic weight. A 2021 report by Oxera found that predictability of legal outcomes directly reduces transaction costs and increases the volume of cross-border trade, showing that the cost and duration of commercial dispute resolution vary sharply across jurisdictions depending on the clarity of applicable rules.22 In the arbitration context specifically, the UK Law Commission found that the uncertainty generated by the Enka principles created ‘unnecessary cost and delay’ through satellite litigation.23 As per LCIA data, these costs can amount to a median cost of USD 117,653 per arbitration even before the substantive dispute is reached.24 The conflict-of-laws analysis required to determine the main contract’s governing law illustrates precisely why that prior question can itself become a source of satellite litigation. In an English court, the governing law of the main contract is determined by the retained Rome I Regulation. 25 Where parties have expressly chosen a governing law under Article 3, no difficulty arises. The problem emerges under Article 4, which applies where no express choice has been made. Courts must first apply the characteristic performance test under Article 4(2), identifying which party’s performance is characteristic of the contract and defaulting to that party’s habitual residence, and 22

Oxera Consulting LLP, ‘The Economic Value of English Law’ (LegalUK, 5 October 2021) <https://legaluk.org/wp-content/uploads/2021/09/The-value-of-English-law-to-the-UK-economy.pdf> accessed 18 June 2026. 23 Review of the Arbitration Act 1996: Final Report and Bill (n 16), para 12.74. 24 London Court of International Arbitration, 'Facts and Figures — Costs and Duration' (30 December 2024) < https://www.lcia.org/media/download.aspx?MediaId=1032> accessed 10 July 2026. 25 Regulation (EC) No 593/2008 of the European Parliament and of the Council on the law applicable to contractual obligations (Rome I) [2008] OJ L177/6, Art 1(2)(e), 2-4; Law Applicable to Contractual Obligations and NonContractual Obligations (Amendment etc) (EU Exit) Regulations 2019, SI 2019/834.

Singapore Comparative Law Review

Converging or Diverging Paths?

then, where the contract is ‘manifestly more closely connected’ with another country, apply the escape clause under Article 4(3). This analysis is fact-sensitive, contestable, and frequently litigated. Enka illustrates this cascade. The main construction contract contained no express governing law clause, so the court applied Rome I Article 4 and concluded that Russian law governed.26 That conclusion fed directly into the Sulamerica Stage 2 implied choice enquiry for the arbitration agreement. Critically, Rome I does not resolve this cascade at the second step. Article 1(2)(e) expressly excludes arbitration agreements from Rome I’s scope, meaning that once the main contract’s governing law has been identified under Articles 3 or 4, the arbitration agreement’s governing law must then be determined separately under common law conflict rules. The result is a two-stage conflict-of-laws analysis: one governed by Rome I (the main contract) and the other governed by common law (arbitration agreement). Where the main contract’s law is contested under Article 4 rather than chosen under Article 3, the Sulamerica/BCY Tier 2 implied-choice inquiry has no stable anchor and the very governing law it is asked to treat as the presumptive choice itself is uncertain. The seat-law default under Section 6A severs this dependency entirely. By making the seat the default regardless of the main contract’s governing law, Section 6A eliminates both stages of the cascade and replaces a layered, fact-sensitive enquiry with a single identifiable answer. The second is separability. Enshrined in Section 7 of the Arbitration Act 1996 and Article 16(1) of the UNCITRAL Model Law, the separability doctrine treats the arbitration agreement as a legally distinct contract from the main agreement in which it appears. Born argues that the arbitration agreement has a fundamentally different character from the underlying commercial obligations it serves.27 It is a procedural compact whose natural governing law is that of the jurisdiction tasked with supervising the arbitration. The seat-law default gives substantive content to separability. It ensures that the arbitration agreement is assessed by the law of the supervisory court, which is best placed to determine its validity and scope. The third is alignment with the New York Convention enforcement architecture. Article V(1)(a) of the Convention provides that recognition and enforcement of an award may be refused where 26

Enka (n 1), [160]. Gary Born, International Commercial Arbitration (3rd edn, Kluwer Law International 2021) vol I, ch 3, s 3.02[B][2], 385; ch 4, s 4.02[A], 512. 27

Singapore Comparative Law Review

Converging or Diverging Paths?

the arbitration agreement is not valid under the law of the country where the award was made (that is, the seat) absent an express governing law designation. Section 6A aligns England’s default with the Convention’s enforcement default, creating coherence between the ex-ante governing law rule and the ex-post enforcement standard. By contrast, SIDRA’s main-contract-law default may produce a mismatch. An arbitration agreement governed by English law (the main contract law) seated in Singapore could be invalid under English law but valid in Singapore law, leaving enforcement courts with conflicting reference points. Finally, the UK seat-law default protects itself from hostile main-contract laws. The seat-law default structurally prevents the problem illustrated by Sulamerica itself, where the main contract’s governing law would negate the arbitration agreement, from arising as a default rule. Where parties have not specified the governing law of the arbitration agreement, applying the seat’s law protects the clause from inadvertent invalidation by a main-contract law that renders certain disputes nonarbitrable or imposes additional formal requirements. Furthermore, the UK, France, Sweden and Germany already apply the seat-law default, and Article 16.4 LCIA Rules adopts the same position. B. Arguments for Singapore’s Main Contract Law Default Meanwhile, the strongest case for Singapore’s main contract law default is fidelity to party expectations. Commercial parties negotiating an international contract with an express governing law clause overwhelmingly intend one legal system to govern the entirety of their contractual relationship. This includes the arbitration clause embedded within it. As Steven Chong J reasoned in BCY, ‘it is reasonable to assume that the contracting parties intend the same system of law to govern all clauses in the contract’. 28 The natural inference is that an express choice of governing law extends to the construction of all clauses in the agreement… including the arbitration agreement. The seat-law default frustrates this expectation by subjecting the arbitration agreement to a law the parties chose only as a procedural venue, not as the substantive law of their obligations. Moreover, the premise underlying the seat-law default, that parties who choose a seat intend its substantive law to govern their arbitration agreement, is empirically unsupported. The

BCY v BCZ (n 11), [59].

Singapore Comparative Law Review

Converging or Diverging Paths?

QMUL/White & Case International Arbitration Survey, the most comprehensive and longitudinal empirical study of user preferences in international arbitration, has consistently found across every edition since 2010 that parties choose a seat for systemic and procedural reasons entirely distinct from that seat’s substantive law. In the 2010 survey, the most important seat-selection factor was ‘formal legal infrastructure’.29 Cited by 62% of respondents, this infrastructure comprised the neutrality and impartiality of the local legal system, the quality of the national arbitration law, and the seat’s track record in enforcing agreements to arbitrate and arbitral awards. The survey’s own authors noted that a substantial minority also cited ‘the law governing the substance of the dispute’ (46%) but expressly flagged this as ‘a rather bizarre outcome, as the law governing the arbitration is typically distinct from the law governing the substance of the dispute’. By 2015, the ‘reputation and recognition of the seat’ had become the dominant factor at 65%, with the three paramount reasons for preferring one seat over another being: ● The neutrality and impartiality of the local legal system. ● The quality of the national arbitration law. ● The track record for enforcing agreements to arbitrate and awards, in that order. 30 The 2021 survey confirmed this hierarchy. When asked what adaptations would make other seats more attractive, respondents ranked greater court support for arbitration (56%), increased neutrality and impartiality (54%), and a better enforcement track record (47%) as the decisive factors. The 2025 survey, the largest ever conducted with 2,402 respondents, confirmed that the factors influencing seat preference remained ‘consistent with those singled out by respondents to our previous surveys’, namely court support, neutrality, the national arbitration law, and enforcement track record.31 This empirical record has a direct doctrinal implication. Parties select Singapore or London as a seat because they trust the institutional infrastructure, the supervisory courts, and the enforcement

Loukas Mistelis, 'Arbitral Seats: Choices and Competition' (Kluwer Arbitration Blog, 26 November 2010) https://legalblogs.wolterskluwer.com/arbitration-blog/arbitral-seats-choices-and-competition/ accessed 16 July 2026; International Arbitration Survey (n 5). 30 International Arbitration Survey (n 5) 27. 31 International Arbitration Survey (n 5) 7.

Singapore Comparative Law Review

Converging or Diverging Paths?

architecture, not because they want that jurisdiction’s substantive contract law to determine the validity or interpretive scope of their arbitration clause. The SIDRA Report’s rationale captures this distinction precisely. Singapore-seated cases typically involve parties and contracts with predominantly foreign elements, and the choice of Singapore as seat reflects confidence in the SIAC, the Singapore courts, and the New York Convention enforcement framework, not a preference for Singapore law as the governing law of the arbitration agreement. Imposing Singapore law on an arbitration agreement by default operation of a seat-law rule inverts the parties’ actual decision-making process. It treats a choice made for procedural and institutional reasons as if it were a substantive choice of law, which is a conflation that the empirical evidence consistently and expressly rejects. Another reason is the interpretive coherence across the contract. Redfern and Hunter describe applying different laws to the main contract and the arbitration agreement as ‘an unreal distinction’,32 noting the practical difficulties that arise when courts might bifurcate their analysis of a single instrument. The American Law Institute’s Restatement of the US Law of International Commercial and Investor-State Arbitration adopts the main-contract law as its default for precisely this reason. Subjecting the arbitration agreement to a different law from the main contract may also generate inconsistency in interpreting related provisions, for instance, where the scope of the arbitration clause tracks defined terms whose meaning is a matter of main contract law. The fourth is that separability has a narrow and vital purpose that does not compel the seat-law default. The doctrine ensures that a challenge to the validity or existence of the main contract does not automatically nullify the arbitration agreement and must be governed by an entirely different law from the main contract in all cases. As the Singaporean courts have consistently observed, separability is a practical device, not a license to impose doctrinal distance between closely related clauses of the same agreement. C. The Unresolved Problem Under Section 6A: What Constitutes ‘Express Agreement’? Both regimes are improvements on what preceded them but Section 6A generates a new interpretive frontier that is worth examining.

Redfern and Hunter (n 6).

Singapore Comparative Law Review

Converging or Diverging Paths?

Section 6A(2) forecloses one source of satellite litigation (the implied-choice analysis) but creates another. The subsection provides that a governing law clause in the main contract ‘does not constitute express agreement’ that that law also governs the arbitration agreement. The question left open is whether a governing law clause framed using defined terms that capture the arbitration agreement as part of ‘this Agreement’ or ‘all obligations arising hereunder’ would nonetheless satisfy Section 6A(1)(a) as an express designation. In Kabab-Ji, the Supreme Court found that the clause ‘this Agreement shall be governed by the laws of England’ did constitute an express choice for the arbitration agreement because the defined term ‘this Agreement’ extended to all clauses. Under Section 6A(2), such a clause ‘does not constitute express agreement’ for the arbitration agreement. However, what of a clause that additionally references ‘all provisions of this Agreement including the dispute resolution provisions’? SIDRA’s framework does not entirely escape this problem either. Its Tier 1 similarly requires parties to ‘expressly designate’ the governing law of the arbitration agreement, and the boundary between express designation and the Tier 2 default will invite litigation of its own. However, what SIDRA does avoid is the additional ambiguity introduced by Section 6A(2)’s attempt to neutralise main-contract governing law clauses since under SIDRA, such clauses operate directly as the Tier 2 default rather than being partially neutered. D. The Article’s Normative Position The article argues that Singapore’s proposed SIDRA framework better serves international commercial parties, subject to one qualification. The primary argument is that Tier 2’s main-contract-law default is the more realistic proxy for party intent in the ordinary case. When commercial parties negotiate a contract governed by English law with a Singapore-seated SIAC clause, they are far more likely to have contemplated the entire contractual relationship (including the arbitration agreement) being governed by English law than to have made a considered choice that the arbitration agreement should be governed by Singapore law as the seat. The SIDRA framework honours that expectation. By defaulting to the law of the seat, Section 6A imposes a governing law on the arbitration agreement that the parties selected for entirely different reasons.

Singapore Comparative Law Review

Converging or Diverging Paths?

The qualification is that Section 6A’s alignment with the New York Convention enforcement architecture is a genuine advantage that should not be dismissed. An arbitration agreement governed by the main contract law but seated in Singapore and enforced under Article V(1)(a) by reference to Singapore law may produce the mismatch problem identified above. If Singapore intends to introduce legislation in this area, the statutory wording eventually enacted should address this directly. Rather than treating the main-contract governing law clause as an irrebuttable Tier 2 default, the enacted provision could treat it as a rebuttable presumption. This would preserve the validation principle’s protective function, which SIDRA’s current formulation inadvertently eliminates in converting BCY’s defeasible implied-choice inference into a hard statutory default. The recommendation, strongly endorsed by Hwang, Uyar and Wimmers, 33 as well as by HKIAC’s approach to model clause design, 34 is that parties should resolve the ambiguity themselves. An express governing-law designation for the arbitration agreement, distinct from the main contract’s governing-law clause, renders the default rule irrelevant under both regimes. This advice remains necessary notwithstanding decades of litigation over this precise question reflects a structural failure of commercial drafting practice that no default rule alone can repair. V.

CONCLUSION

Both Section 6A and SIDRA represent genuine advances over the Sulamerica framework. The doctrinal uncertainty that generated Enka, Kabab-Ji, BNA, and Anupam Mittal (and the costs that those cases imposed on parties attempting to resolve threshold jurisdictional questions) is substantially reduced under both regimes. In that vein, the reforms should be welcomed. However, the reforms create a divergence between two of the world’s most popular seats. Commercial parties drafting arbitration clauses for contracts with cross-jurisdictional elements (English governing law, Singapore seated) now operate in a landscape where two plausible supervisory jurisdictions would reach opposite conclusions on the governing law of their

Michael Hwang SC, Gökçe Uyar and Cosima Wimmers, 'The Enka v Chubb/Anupam Mittal v Westbridge Controversies: Why Not the Hong Kong (Partial) Solution?' (2024) 41(5) Journal of International Arbitration 41 (Issue 5) 531. 34 Hong Kong International Arbitration Centre (HKIAC), Administered Arbitration Rules (2024) Model Arbitration Clause.

Singapore Comparative Law Review

Converging or Diverging Paths?

arbitration agreement, producing a new form of uncertainty in the cross-border transactions that international commercial arbitration exists to serve. The more important insight, which the reforms invite but do not themselves provide, concerns the direction of thought leadership going forward. If both regimes agree that express designation trumps any default, and if virtually all commentators agree that express designation is advisable, the productive normative question is no longer which default is better, but why parties fail to specify the governing law of the arbitration agreement in the first place. The answer is partly a failure of drafting checklists and partly a tendency to treat the arbitration clause as boilerplate rather than as a substantive negotiated provision. The appropriate response from institutions, practitioners, and scholars is to embed express governing law designations in standard model arbitration clauses and to require them as a condition of institutional registration. The HKIAC model clause already moves in this direction, and perhaps Singapore can follow suit in revising its institutional guidance on model clause design.

About Us Local Strength. Global Excellence. Exceptional Opportunities. Baker McKenzie.Wong & Leow is a joint law venture between Baker McKenzie and Wong & Leow LLC. Combining a global platform with Singapore law capability, we advise on domestic, regional and cross-

Law Firm Brand

World’s strongest law firm brand for 12 years running (Thomson Reuters)

border matters through one integrated team. Through Wong & Leow LLC, we have full right of audience before the Singapore courts. Our locally admitted and internationally qualified lawyers combine deep local knowledge with a global perspective to deliver practical, commercially focused solutions. For aspiring lawyers, this means gaining exposure to both local and

Our Practice Groups Capital Markets | Dispute Resolution Employment | Finance & Projects

international work, learning from leading practitioners, and building

Financial Services Regulatory

a career within a collaborative and globally connected firm.

IPTech | Mergers & Acquisitions Private Equity Funds | Real Estate | Tax

Culture of Friendship At the heart of Baker McKenzie is a culture of friendship that has been nurtured for more than 75 years. We value respectful, collaborative and community-minded behaviours, fostering an environment where people support one another and grow professionally together. We are equally committed to making a positive impact through our pro bono and community initiatives, including partnerships with the

START YOUR JOURNEY WITH US. Scan to explore our Internship and Traineeship opportunities.

Yellow Ribbon Fund, Beyond Social Services, ISCOS and Global Rights for Women, as well as volunteering at the Law Society’s Legal Aid Clinic.

Realise Your Potential We are committed to helping our people realise their full potential

Scan for the latest news and updates.

Connect with us now!

through structured training, coaching and mentoring, empowering them with the support and opportunities to grow at every stage of their careers. From working on local and cross-border matters to international and client secondment opportunities, you will gain the experience, skills and global perspective needed to thrive in an increasingly interconnected world.

© 2026 Baker & McKenzie.Wong & Leow is a member firm of Baker & McKenzie International, a global law firm with member law firms around the world. In accordance with the common terminology used in professional service organizations, reference to a “partner” means a person who is a partner or equivalent in such a law firm. Similarly, reference to an “office” means an office of any such law firm. This may qualify as “Attorney Advertising” requiring notice in some jurisdictions. Prior results do not guarantee a similar outcome.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

RETHINKING THE PERSON SKILLED IN THE ART IN THE AGE OF ARTIFICIAL INTELLIGENCE: INVENTIVE STEP IN SINGAPORE, UK AND EUROPEAN PATENT LAW Joia Ng Yi Jing* Runner Up for Best Article Award 2026 ABSTRACT Artificial intelligence is increasingly embedded in research across pharmaceuticals, materials science and advanced engineering. This development places pressure on patent law’s conception of the person skilled in the art (“PHOSITA”), the hypothetical benchmark used to assess inventive step. The difficulty is not that the benchmark remains human, but that the knowledge, resources and working methods attributed to it may no longer reflect ordinary research practice. A skilled person assessed without tools that researchers routinely use may understate ordinary technical capacity, allowing standard AI-assisted work to appear inventive. The opposite approach is equally problematic. Attributing frontier AI capabilities to the skilled person may cause results that remained difficult for ordinary researchers at the priority date to be treated as obvious. This article therefore proposes an AI-augmented skilled person grounded in both common general knowledge (“CGK”) and the ordinary technical means available in the field. The benchmark remains a human practitioner or team. An AI tool should affect the inventive-step baseline only where ordinary practitioners could realistically access it, regard it as a reliable basis for the relevant task, and deploy it without exceptional expertise, data or infrastructure. Drawing on the patent law of Singapore and the United Kingdom (“UK”), together with the doctrine and examination practice of the European Patent Office (“EPO”), the article argues that existing principles of CGK, ordinary technical capacity and inventive step can accommodate this adjustment. Statutory reform is unnecessary. Singapore is particularly well placed to develop the approach through guidance issued by the Intellectual Property Office of Singapore (“IPOS”) and

Singapore Management University (SMU) (LLM class of 2025)

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

through judicial reasoning because its patent law draws closely on UK jurisprudence and its innovation policy is framed by the Singapore IP Strategy 2030. I. A.

INTRODUCTION

The Problem: AI in Routine Research

In 2020, DeepMind's AlphaFold produced protein-structure predictions at a level of accuracy widely described as transformative for structural biology, predicting three-dimensional structures directly from amino acid sequences with a degree of reliability that had eluded researchers for decades. 1 AlphaFold and its successors have since generated hundreds of millions of structural predictions

at unprecedented scale, accelerating structural-biology and drug-discovery workflows and extending AI’s relevance beyond single-protein prediction to broader biomolecular interactions involving proteins, DNA, RNA, ligands and small molecules.2 AlphaFold is not an isolated case. Across drug discovery, materials science, chemical engineering, and semiconductor design, AI systems are increasingly used to generate, filter, refine, and validate technical solutions at a scale that no unaided human team could independently match. 3 The legal framework through which the resulting patent claims are assessed, however, was developed for a world in which such tools did not exist, and it has not yet fully absorbed the doctrinal implications of their widespread adoption. 4

John Jumper and others, ‘Highly Accurate Protein Structure Prediction with AlphaFold’ (2021) 596 Nature 583, 583–84. 2 Mihaly Varadi and others, ‘AlphaFold Protein Structure Database: Massively Expanding the Structural Coverage of Protein-Sequence Space with High-Accuracy Models’ (2022) 50 Nucleic Acids Research D439, D440; Mihaly Varadi and others, ‘AlphaFold Protein Structure Database in 2024: Providing Structure Coverage for Over 214 Million Protein Sequences’ (2024) 52 Nucleic Acids Research D368, D368; AlphaFold Protein Structure Database, ‘AlphaFold DB’ (EMBL-EBI) <https://alphafold.ebi.ac.uk/> accessed 4 July 2026; Josh Abramson and others, ‘Accurate Structure Prediction of Biomolecular Interactions with AlphaFold 3’ (2024) 630 Nature 493, 493–94. 3 Fabio Urbina and others, ‘Dual Use of Artificial-Intelligence-Powered Drug Discovery’ (2022) 4 Nature Machine Intelligence 189, 189–90; Noam Shemtov, A Study on Inventorship in Inventions Involving AI Activity (European Patent Office 2019) 7–9; Daniel C Elton and others, ‘Deep Learning for Molecular Design—A Review of the State of the Art’ (2019) 4 Molecular Systems Design & Engineering 828, 828–49; Keith T Butler and others, ‘Machine Learning for Molecular and Materials Science’ (2018) 559 Nature 547, 547–55; Azalia Mirhoseini and others, ‘A Graph Placement Methodology for Fast Chip Design’ (2021) 594 Nature 207, 207–12. 4 United States Patent and Trademark Office, Public Views on Artificial Intelligence and Intellectual Property Policy (October 2020) 15; Noam Shemtov and Garry A Gabison, ‘The Inventive Step Requirement and the Rise of

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

PHOSITA as the Calibration Device

Patent law evaluates inventive step through the perspective of the person skilled in the art. This hypothetical construct, commonly referred to as PHOSITA, denotes a human technician of ordinary skill in the relevant technical field, familiar with the common general knowledge (“CGK”) of that field but lacking creative or inventive ingenuity. 5 It is not merely a convenient fiction, but the calibration device for the entire obviousness inquiry.6 An invention that would have been obvious to the skilled person does not merit patent protection, whereas an invention that goes beyond what such a person would achieve through routine skill and knowledge may justify a temporary monopoly as an incentive for disclosure and investment in research. 7 The way PHOSITA is defined therefore determines how demanding the patent system’s quality filter is in practice.8 C.

Two Symmetrical Risks

The integration of AI into routine research creates two symmetrical risks if PHOSITA is left conceptually unchanged. If the skilled person is understood as a strictly human practitioner without access to widely used AI tools, patent offices and courts risk granting patents for outputs that AI systems can routinely generate from publicly available data and standard methods. 9 In AI-intensive biomedical research, that risk echoes the anticommons dynamic identified by Heller and Eisenberg, in which overlapping upstream rights over routine discoveries may impede downstream innovation and fragment cumulative research programmes. 10 It is also consistent with recent

the AI Machines’ in Ryan Abbott (ed), Research Handbook on Intellectual Property and Artificial Intelligence (Edward Elgar 2022) 423–42. 5 Windsurfing International Inc v Tabur Marine (Great Britain) Ltd [1985] RPC 59, 73–74 (Oliver LJ); Pozzoli SPA v BDMO SA [2007] EWCA Civ 588, [2007] FSR 37, [23] (Jacob LJ). 6 Windsurfing (n 5) 73–74; Pozzoli (n 5) [23]. 7 Biogen Inc v Medeva plc [1997] RPC 1, 34 (Lord Hoffmann); Actavis Group PTC EHF v ICOS Corporation [2019] UKSC 15, [2019] Bus LR 1318, [17], [59]–[60] (Lord Hodge). 8 Actavis (n 7) [59]–[60]; Suzanne Scotchmer, ‘Standing on the Shoulders of Giants: Cumulative Research and the Patent Law’ (1991) 5 Journal of Economic Perspectives 29, 30. 9 Ryan Abbott, ‘Everything Is Obvious’ (2019) 66 UCLA Law Review 2, 8–11; Maurice Schellekens, ‘Artificial Intelligence and the Re-Imagination of Inventive Step’ (2022) 13(2) JIPITEC 89, 91–97. 10 Michael A Heller and Rebecca S Eisenberg, ‘Can Patents Deter Innovation? The Anticommons in Biomedical Research’ (1998) 280 Science 698, 698–701.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

empirical concerns that AI-native pharmaceutical firms may be able to reach patent-facing outputs with less traditional pre-clinical validation than conventional developers. 11 The opposite risk arises if PHOSITA is calibrated by reference to frontier AI capability rather than ordinary research practice. A claimed solution might then be treated as obvious merely because an advanced AI system could identify it, even though the same result remained beyond what ordinary practitioners could achieve through routine work at the priority date.12 This would raise the inventive-step threshold beyond ordinary skill and risk denying protection across AI-intensive fields.13 Neither extreme offers a satisfactory account of the skilled-person inquiry. This article therefore adopts a middle position under which PHOSITA should not be reconceived as AI itself, but neither should it remain artificially detached from AI tools that ordinary practitioners routinely use in the field.14 The unresolved doctrinal question is therefore when the skilled person should be attributed with knowledge of, practical access to and ordinary competence in using a particular AI tool. The inquiry should ask whether the tool was accessible in practice, routinely used and accepted as reliable for the relevant technical purpose, and deployable through ordinary professional competence at the priority date.15 D.

Positioning Within the Literature

The scholarship on AI and patent law has, for understandable reasons, concentrated primarily on inventorship. The DABUS litigation in the UK culminated in the Supreme Court’s decision in Thaler, which held that under the Patents Act 1977, an inventor must be a natural person and that recognising an AI system as inventor would require legislative change rather than judicial

Janet Freilich and Arti K Rai, ‘What Patents on AI-Derived Drugs Reveal’ (2025) 388 Science 924, 924–25. Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Schellekens (n 9) 95–97. 13 Ryan Abbott, ‘I Think, Therefore I Invent: Creative Computers and the Future of Patent Law’ (2016) 57 Boston College Law Review 1079, 1110–13. 14 Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49, [2024] AC 1, [56]–[65], [73] (Lord Kitchin); Abbott, ‘I Think, Therefore I Invent’ (n 13) 1105–13. 15 General Tire & Rubber Co v Firestone Tyre & Rubber Co Ltd [1972] RPC 457, 482–83 (Sachs LJ); First Currency Choice Pte Ltd v Main-Line Corporate Holdings Ltd [2007] SGCA 50, [2008] 1 SLR(R) 335, [28], [38]– [44]; Mühlbauer AG v Manufacturing Integration Technology Ltd [2010] SGCA 6, [2010] 2 SLR 724, [20]–[21]; Intellectual Property Office of Singapore, Examination Guidelines for Patent Applications at IPOS (October 2023) paras 2.24–2.32; European Patent Office, Guidelines for Examination in the European Patent Office (April 2026) GVII, 3. 12

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

innovation.16 In Singapore, Saw and Chan have examined AI inventorship and ownership under the existing Patents Act framework, concluding that current provisions can accommodate present levels of AI assistance while acknowledging that future reform may be needed if fully autonomous AI inventors emerge.17 Abbott’s earlier work on creative computers canvassed the normative implications of machine creativity for patent incentives, while Shemtov’s empirical study for the EPO mapped how AI is already used in inventive processes across a range of technical sectors.18 Recent scholarship has begun to examine how ordinary use of AI bears on the skilled-person inquiry. Schellekens and Shemtov and Gabison address inventive step directly. Kahwaji considers the effect of AI tools on the skilled person’s general knowledge and ordinary working environment, while Bulat argues that the benchmark should be understood as “AI-equipped”.19 This emerging literature confirms that inventive step, rather than inventorship alone, now requires sustained attention. The present article develops the debate through a Singapore-centred comparison with the UK and the EPO. It also distinguishes what the skilled person ordinarily knows from the resources and capacities available in practice. On that basis, it develops an evidential method for deciding when a particular AI tool should affect the inventive-step baseline and explains how IPOS could apply that method.20 In many current research settings, AI is used as a tool: a human researcher identifies the problem, selects and deploys an AI system, and interprets the outputs in light of domain expertise and experimental validation. 21 In such cases, the harder issue is not inventorship, but whether routine AI use should alter the baseline for non-obviousness. This article addresses that question.

Thaler (n 14) [56]–[65], [73]. Cheng Lim Saw and Samuel Zheng Wen Chan, ‘Of Inventorship and Patent Ownership: Examining the Intersection between Artificial Intelligence and Patent Law’ [2023] Sing JLS 27, 27–28, 34–35. 18 Abbott, ‘I Think, Therefore I Invent’ (n 13) 1105–13; Shemtov (n 3) 7–9. 19 Schellekens (n 9) 91–97; Shemtov and Gabison (n 4) 423–42; Alia Kahwaji, ‘Safeguarding a Human-Centric Patent System: The Case of the Inventive Step Test’ (2025) 16(1) European Journal of Law and Technology <https://ejlt.org/index.php/ejlt/article/view/1086> accessed 10 July 2026; Nataliia Bulat, ‘When It Comes to the Use of AI in the Inventive Process: Is There Room to Confirm the Subjective Inventorship Standard in a World of Objective Patentability Criteria?’ (2026) Journal of World Intellectual Property <https://doi.org/10.1111/jwip.70019> accessed 10 July 2026. 20 The distinctions and evidential questions developed in this article are the author’s synthesis. See General Tire (n 15) 482–83; First Currency Choice (n 15) [28], [38]–[44]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 21 Elton and others (n 3) 828–49; Shemtov (n 3) 7–9. 17

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

Scope and Limitations

Two limitations should be stated at the outset. The argument is directed to Level 1 AI-as-tool cases, in which a human researcher deploys a practically accessible AI tool and exercises independent judgment over the outputs. 22 It does not attempt to resolve the harder problem at the Level 1/2 boundary, where AI contributes substantially to generating the inventive concept itself. 23 The argument is also field-contingent and priority-date specific. The proposed evidential questions do not create a universal rule that every AI tool forms part of the skilled person’s ordinary technical environment.24 They instead provide a framework for deciding whether a particular tool should affect the benchmark in a particular field at the relevant date.25 F.

Why Singapore, the UK, and the EPO

This comparative inquiry is particularly relevant to Singapore. Singapore’s patent law is closely aligned with UK jurisprudence, and Singapore courts have treated UK patent authorities as highly persuasive in interpreting the inventive-step requirement under section 15 of the Patents Act. 26 IPOS examination practice is also informed by both UK and EPO materials, including the Windsurfing/Pozzoli approach to obviousness and the EPO’s problem-solution method.27 The Singapore-UK-EPO comparison therefore provides a practical framework for considering how Singapore might respond to AI-assisted inventive step without departing from established doctrine.28 The EPO materials are especially useful because EPO Guidelines and Board of Appeal decisions already contain detailed reasoning on computer-implemented inventions, machine-learning methods, and the skilled person’s ordinary technical capacities. 29 The EPO Guidelines describe

Shemtov (n 3) 7–9; Elton and others (n 3) 828–49. Freilich and Rai (n 11) 924–25. 24 General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 25 EPO, Guidelines (n 15) G-VII, 3. 26 First Currency Choice (n 15) [35]–[44]; Mühlbauer (n 15) [18]–[21]; Ng-Loy Wee Loon, Law of Intellectual Property of Singapore (3rd edn, Sweet & Maxwell 2021) paras 30.2.8–30.2.15. 27 IPOS, Examination Guidelines (n 15) paras 2.24–2.32, 4.18–4.24; Intellectual Property Office of Singapore, Supplemental Guidance for the Examination of AI-Related Patent Applications (October 2024) paras 1–5, 15–17. 28 First Currency Choice (n 15) [35]–[44]; IPOS, Examination Guidelines (n 15) paras 4.18–4.24. 29 EPO, Guidelines (n 15) G-II, 3.3.1, G-VII, 3; T 0702/20 Sparsely Connected Neural Network/Mitsubishi (EPO Technical Board of Appeal, 7 November 2022) [14]–[19]; T 1952/21 Reinforcement Learning/Bosch (EPO Technical Board of Appeal, 14 June 2024) [22]–[24.1]. 23

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

the skilled person as having access to the means and capacity for routine work and experimentation that are normal for the field, a formulation that is particularly relevant when AI tools become routine research instruments. 30 Recent Board of Appeal decisions also show the limits of the analysis: AI and machine-learning techniques may be relevant to technical contribution and inventive step, but AI-generated material cannot simply substitute for the skilled person’s understanding.31 G.

Thesis and Roadmap

The central question is whether PHOSITA remains an adequate benchmark for assessing inventive step when ordinary research practice in the relevant field includes the routine use of AI tools. The human nature of the benchmark is not the problem. The difficulty arises when the knowledge, resources and working methods attributed to the skilled person no longer reflect ordinary practice in the relevant field.32 An inventive-step inquiry that excludes tools already embedded in ordinary research practice risks understating the capabilities of the skilled person, with the result that routine AI-assisted work may be treated as inventive. The converse is equally problematic. If the skilled person is assumed to possess the capabilities of frontier AI, results that remain beyond what ordinary researchers could reasonably achieve at the priority date may instead be regarded as obvious.33 The skilled person should remain a human practitioner or, where appropriate, a team. At the same time, the tools attributed to that benchmark must be capable of reflecting changes in ordinary research practice. Whether a particular AI tool should alter the inventive-step baseline will depend on how it was used in the relevant field at the priority date. The inquiry should consider whether ordinary practitioners could realistically access the tool, whether they regarded it as reliable for the task in question, and whether they could use it without exceptional expertise or resources.34

EPO, Guidelines (n 15) G-VII, 3. T 1193/23 (EPO Technical Board of Appeal, 15 April 2025) catchword, [1.1.1]; T 0702/20 (n 29) [19]; T 1952/21 (n 29) [22]–[24.1]. 32 General Tire (n 15) 482–83; First Currency Choice (n 15) [28], [41]–[44]; Mühlbauer (n 15) [20]–[21]. 33 Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Schellekens (n 9) 93–97. 34 General Tire (n 15) 482–83; First Currency Choice (n 15) [28], [41]–[44]; EPO, Guidelines (n 15) G-VII, 3. 31

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

This does not equate AI with the skilled person or make AI-assisted inventions presumptively obvious. It asks whether the tool had become part of the ordinary way in which practitioners approached the relevant technical problem. Even where that is established, the tribunal must still decide whether the claimed invention itself would have been obvious. 35 Part II sets out the comparative doctrinal foundations in Singapore, the United Kingdom and the European Patent Office. Part III examines how those systems construct the skilled person and assess ordinary technical practice. Part IV considers the pressures created by AI-assisted research. Part V develops the proposed framework. Part VI considers its application and significance for Singapore. Part VII concludes. II.

DOCTRINAL FOUNDATIONS OF INVENTIVE STEP A.

The Function of Inventive Step

The inventive step requirement is the patent system’s principal quality filter, giving practical effect to the patent bargain by conditioning the grant of exclusive rights on a contribution that goes materially beyond what an ordinarily skilled person could have achieved at the priority date through routine knowledge and standard skill. 36 The normative stakes of calibrating that threshold correctly are well recognised: a threshold set too low protects incremental advances that would have emerged in the ordinary course of research, withdrawing subject matter from the public domain and raising the cost of follow-on innovation, while a threshold set too high denies protection to genuine advances, weakening the incentive to invest in expensive and uncertain research programmes.37 These consequences are not merely theoretical. The height of the obviousness threshold affects what kinds of research are rewarded, what claims may impede later work, and how confidently firms can invest in long-horizon development.38 For that reason, the balance between over- and under-protection cannot be struck in the abstract. It is mediated through the perspective of the

Thaler (n 14) [56]–[65], [73]; T 1193/23 (n 31) [1.1.1]. Windsurfing (n 5) 73–74; Biogen (n 7) 34. 37 Heller and Eisenberg (n 10) 698–701; Scotchmer (n 8) 30. 38 Heller and Eisenberg (n 10) 698–701; Scotchmer (n 8) 30. 36

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

person skilled in the art, whose knowledge, capacities, and ordinary research methods define the baseline against which inventive contribution is judged.39 As the UK Supreme Court’s reasoning in Actavis illustrates, obviousness requires an evaluative assessment of what the skilled person would have understood, what the CGK of the field was, and what routine steps such a person would have been prepared to take at the priority date. 40 The obviousness inquiry therefore turns not only on the claim itself but also on how the law constructs the notional benchmark against which the claim is measured. 41 If that benchmark is misdescribed, the analysis is distorted from the outset rather than at the margins.42 This is precisely the risk in AI-intensive fields: where the legal construct of the skilled person no longer reflects ordinary research practice, the inventive-step threshold will be miscalibrated from the beginning.43 That is not because patent law lacks the concepts to respond to technological change, but because those concepts, particularly CGK, may be applied too mechanically unless courts and patent offices are willing to ask what ordinary technical work now actually looks like.44 B.

The Skilled Person

Notwithstanding its deceptively simple label, the person skilled in the art is a legal construct of considerable sophistication: not any actual scientist, engineer, or technician, but a notional benchmark used to determine whether the claimed invention rises above the level of ordinary technical work.45 The character attributed to the skilled person determines how demanding the obviousness inquiry will be in practice, which is why the courts have developed this construct with care over several decades. 46 This construct matters for AI-assisted invention because the present issue is not whether the skilled person should become more inventive or more technologically powerful in the abstract. 47 It is 39

Actavis (n 7) [59]–[60]; First Currency Choice (n 15) [41]–[44]. Actavis (n 7) [59]–[60], [88]. 41 Pozzoli (n 5) [23]; First Currency Choice (n 15) [41]–[44]. 42 First Currency Choice (n 15) [44]–[45]. 43 Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Schellekens (n 9) 91–97. 44 General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) GVII, 3. 45 Pozzoli (n 5) [23]; Mühlbauer (n 15) [21]. 46 First Currency Choice (n 15) [28]–[31], [41]–[44]; Mühlbauer (n 15) [20]–[21]. 47 Abbott, ‘I Think, Therefore I Invent’ (n 13) 1105–13. 40

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

whether the ordinary skill attributed to that person should reflect the tools and methods that ordinary practitioners in the relevant field actually use.48 If AI tools have become part of routine technical work, excluding them from the skilled person’s assumed toolkit risks understating ordinary competence and mis-calibrating the inventive-step threshold.49 The skilled person possesses ordinary skill and average ability in the relevant field. 50 As Jacob LJ put it in Pozzoli, they are a skilled but unimaginative worker, technically proficient and familiar with the working knowledge of the field, capable of performing routine research tasks with competence, but not assumed to possess the spark of invention itself. 51 Inventive step is therefore measured against ordinary technical capacity rather than exceptional brilliance. 52 If the benchmark were pitched too high, courts would collapse the line between routine development and genuine invention, and if pitched too low, patent law would reward outcomes representing no more than standard professional work.53 The difficulty posed by AI is not that it displaces the human benchmark. It is that, in some fields, the ordinary competence attributed to that benchmark may now include the routine use of AI tools.54 Patent law can accommodate this development in principle, as the skilled person is already assessed by reference to the relevant field and the relevant priority date. 55 The real difficulty is evidential. A court or patent office must determine, on the evidence, whether a particular AI tool was sufficiently established in the field at the priority date to form part of the skilled person’s ordinary working toolkit.56 The skilled person may also be a team rather than a single individual. 57 In Actavis, the UK Supreme Court confirmed that in technologically complex fields such as pharmaceuticals and biotechnology, the notional addressee may consist of a composite team whose members

Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Schellekens (n 9) 91–97. Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Freilich and Rai (n 11) 924–25. 50 EPO, Guidelines (n 15) G-VII, 3; Mühlbauer (n 15) [21]. 51 Pozzoli (n 5) [23]; Mühlbauer (n 15) [21]. 52 Actavis (n 7) [59]–[60]; Pozzoli (n 5) [23]. 53 Biogen (n 7) 34; Actavis (n 7) [59]–[60]. 54 Shemtov (n 3) 7–9; Elton and others (n 3) 828–49; Butler and others (n 3) 547–55. 55 First Currency Choice (n 15) [41]–[44], [50]; Mühlbauer (n 15) [20]–[21]; EPO, Guidelines (n 15) G-VII, 3. 56 General Tire (n 15) 482–83; First Currency Choice (n 15) [50]; IPOS, Examination Guidelines (n 15) paras 2.24– 2.32. 57 Actavis (n 7) [59]. 49

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

collectively possess the ordinary expertise needed for the task at hand. 58 That aspect of the doctrine matters considerably for AI-assisted invention, since a team-based PHOSITA can naturally accommodate ordinary competence in data science, modelling, or AI-tool deployment where those skills have become standard in the relevant field.59 A modern drug-discovery team, for instance, may include members capable of operating standard computational screening tools as part of ordinary research work.60 Existing doctrine is therefore more flexible than it might first appear: it permits the law to attribute multidisciplinary ordinary competence to the skilled person without abandoning the human-centred nature of the construct. 61 C.

Common General Knowledge

The skilled person’s knowledge is defined in large part by their CGK, which the Court of Appeal in General Tire described as information that is generally known and generally regarded by the bulk of those engaged in the particular art as a good basis for further action. 62 The formulation contains two distinct ideas. The information must be sufficiently general within the relevant practitioner community, and it must be accepted as a reliable foundation for further work rather than remain merely theoretical or contested. 63 That definition addresses what the skilled person knows; it does not, by itself, describe every resource available in practice. For AI tools, the distinction is important. CGK bears on whether practitioners knew of the tool, understood its established uses and regarded it as a sound basis for the technical task in question. 64 Practical access, computing infrastructure and the competence needed to deploy the tool belong to the wider construction of the skilled person’s ordinary technical environment.65 Public availability is therefore insufficient on either account. A model may be

Actavis (n 7) [59]–[60]. Actavis (n 7) [59]–[60]; Elton and others (n 3) 828–49. 60 Elton and others (n 3) 828–49; Butler and others (n 3) 547–55; Freilich and Rai (n 11) 924–25. 61 Actavis (n 7) [59]–[60]; Thaler (n 14) [56]–[65], [73]. 62 General Tire (n 15) 482–83; Halliburton Energy Services Inc v Smith International (North Sea) Ltd [2006] EWCA Civ 1715, [2007] RPC 25, [13]. 63 General Tire (n 15) 482–83; Halliburton (n 62) [13]. 64 General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 65 First Currency Choice (n 15) [28], [31]; EPO, Guidelines (n 15) G-VII, 3. 59

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

discoverable without being accepted in practice, or well known without being realistically usable by an ordinary practitioner or team. 66 Singapore authority confirms that the skilled person must be identified at the relevant priority date and attributed the ordinary competence of the technical field. In Mühlbauer, the Court of Appeal restated the Windsurfing framework and described the notional addressee as ordinarily skilled, competent and unimaginative. 67 The decision did not itself require the Court to determine the content of CGK, and it should not be treated as deciding whether particular tools or working methods form part of ordinary practice. The relevance of Mühlbauer to the present argument is therefore limited but important. It confirms that obviousness is assessed through an ordinary skilled person rather than an exceptionally capable expert.68 The more specific question of whether a technical method or tool was generally known and accepted as a basis for further action must be answered through the established CGK principles reflected in General Tire, the IPOS Examination Guidelines and appropriate expert evidence in the relevant field. Those materials help determine what practitioners knew and accepted, while the wider skilled-person inquiry addresses the technical means and competence ordinarily available to them. The capacity of the skilled-person inquiry to reflect developments in technical practice provides an important doctrinal basis for the present argument. Molecular-modelling programs were once largely confined to specialist computational-chemistry groups, but later became part of routine practice in areas of organic chemistry and pharmaceutical research as access widened and their use became more established. 69 The point is not that any court has formally declared a named software

General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) GVII, 3. The distinction between discoverability, accepted use and realistic deployability is the author’s analysis. 67 Mühlbauer (n 15) [20]–[21]. 68 Mühlbauer (n 15) [20]–[21]; General Tire (n 15) 482–83; First Currency Choice (n 15) [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 69 Edoardo Aprà and others, ‘NWChem: Past, Present, and Future’ (2020) 152 Journal of Chemical Physics 184102; Leopold Talirz, Luca M Ghiringhelli and Berend Smit, ‘Trends in Atomistic Simulation Software Usage [Article v1.0]’ (2021) 3(1) Living Journal of Computational Molecular Science 1483.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

package to form part of CGK. 70 Rather, patent law is capable of recognising that technical methods and tools may move from specialist use into ordinary practice as the field develops. 71 A comparable process is now taking place with AI tools, although adoption differs sharply across fields and over time.72 The legal question is not whether the skilled person can evolve with technical practice. Existing doctrine already identifies that benchmark by reference to the relevant field and priority date. 73 The harder question is evidential. A court must decide when a particular tool became sufficiently known, accessible and relied upon for a defined technical purpose to form part of the ordinary working environment of the relevant practitioner or team. 74 D.

The Structured Tests and the Doctrinal Bridge

Singapore and UK: The Windsurfing/Pozzoli Framework The four-step framework in Windsurfing and Pozzoli provides the structured method through which Singapore and UK law assess inventive step. 75 In its modern form, the inquiry first identifies the notional person skilled in the art and their relevant CGK; secondly, identifies the inventive concept of the claim; thirdly, identifies the differences between the prior art and the inventive concept; and fourthly, asks whether those differences would have been obvious to the skilled person or required inventive ingenuity. 76 Steps (1)(a) and (1)(b) are the gateway to the entire inquiry.77 If the skilled person or their CGK is misdescribed at the outset, the error is not easily corrected at the final obviousness stage, because every later step proceeds on the baseline already fixed. 78 This is why the structured approach is more than a matter of analytical tidiness. 79 It helps discipline the court against hindsight by

General Tire (n 15) 482–83. General Tire (n 15) 482–83; EPO, Guidelines (n 15) G-VII, 3. 72 Shemtov (n 3) 7–9; Elton and others (n 3) 828–49; Butler and others (n 3) 547–55; Freilich and Rai (n 11) 924– 25. 73 General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 74 General Tire (n 15) 482–83; EPO, Guidelines (n 15) G-VII, 3. 75 Windsurfing (n 5) 73–74; Pozzoli (n 5) [23]; First Currency Choice (n 15) [41]–[44]. 76 Pozzoli (n 5) [23]; Windsurfing (n 5) 73–74. 77 Pozzoli (n 5) [23]. 78 First Currency Choice (n 15) [44]–[45]; Mühlbauer (n 15) [20]–[21]. 79 First Currency Choice (n 15) [44]–[45]. 71

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

requiring the tribunal to identify the skilled person and their CGK before asking whether the claimed invention was obvious. 80 The hindsight concern is not merely theoretical. Mandel’s empirical work on cognitive bias in obviousness determinations suggests that decision-makers who already know the claimed solution may unconsciously reconstruct the prior art and the skilled person’s assumed knowledge in ways that make the solution appear more foreseeable than it would have appeared at the priority date. 81 The same risk arises in AI-assisted inventions, although in a more technologically specific form. 82 Once the tribunal knows that AI contributed to the result, that knowledge may distort the body of knowledge, methods and technical resources attributed to the skilled person.83 The distortion may operate in either direction. If AI tools are wrongly omitted from the skilled person’s ordinary technical environment, the skilled person might be artificially weakened, and the inventive-step threshold may become too low. 84 If frontier AI capability is too readily attributed to the skilled person, merely because an advanced system could have produced the result may instead make the benchmark unrealistically powerful.85 The central task is therefore to fix the skilled person’s ordinary toolkit at the priority date, rather than reason backwards from the fact that AI was involved in the inventive process. 86 Singapore courts have applied this structured approach consistently, though without yet confronting the AI-specific version of this hindsight problem directly. In First Currency Choice, the Court of Appeal accepted the continuing usefulness of the Windsurfing framework and emphasised the discipline and consistency it brings to obviousness analysis. 87 The court also recognised that the first three steps lay the groundwork for the final question of whether the alleged

First Currency Choice (n 15) [44]–[45]; Mühlbauer (n 15) [20]. Gregory N Mandel, ‘Patently Non-Obvious: Empirical Demonstration that the Hindsight Bias Renders Patent Decisions Irrational’ (2006) 67 Ohio State Law Journal 1391, 1392–95, 1421–23. 82 Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Schellekens (n 9) 91–97. 83 Mandel (n 81) 1392–95, 1421–23; Abbott, ‘Everything Is Obvious’ (n 9) 8–11. 84 Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Freilich and Rai (n 11) 924–25. 85 Abbott, ‘I Think, Therefore I Invent’ (n 13) 1110–13; Schellekens (n 9) 95–97. 86 First Currency Choice (n 15) [44]–[45]; Mühlbauer (n 15) [20]–[21]. 87 First Currency Choice (n 15) [41]–[44]. 81

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

invention is obvious.88 Mühlbauer later restated the structure and reinforced the need to assume the position of the skilled but unimaginative addressee at the priority date. 89 That sequencing discipline is central to the proposal advanced here. Identifying the skilled person’s CGK and ordinary technical means before addressing obviousness reduces the risk that either will be reconstructed after the result is known.90 The proposed AI-augmented PHOSITA therefore does not modify the Windsurfing/Pozzoli framework.91 It instead refines the factual baseline fixed at the first stage of the inquiry by requiring the tribunal to identify the skilled person’s ordinary knowledge and technical means before assessing obviousness.92 The EPO: The Problem-Solution Approach The EPO’s problem-solution approach structures the inventive-step inquiry differently, but it shares the same foundational commitment to the skilled person as the calibration mechanism. 93 The approach requires identification of the closest prior art, determination of the objective technical problem, and assessment of whether the skilled person, starting from the closest prior art and the objective technical problem, would have arrived at the claimed invention in an obvious way.94 What matters for present purposes is not merely the structure of the test, but where it locates its own flexibility.95 The EPO Guidelines describe the skilled person as having average knowledge and ability, awareness of what was CGK in the art at the relevant date, and access to the means and capacity for routine work and experimentation that are normal for the field. 96 That final phrase does significant doctrinal work,97 because it ties the skilled person’s assumed resources to what counts as normal practice in the relevant technical field at the relevant time. 98

First Currency Choice (n 15) [44]. Mühlbauer (n 15) [20]–[21]. 90 First Currency Choice (n 15) [44]–[45]; Mandel (n 81) 1421–23. 91 Windsurfing (n 5) 73–74; Pozzoli (n 5) [23]. 92 General Tire (n 15) 482–83; Mühlbauer (n 15) [20]–[21]. 93 EPO, Guidelines (n 15) G-VII, 5. 94 EPO, Guidelines (n 15) G-VII, 5.1–5.3. 95 EPO, Guidelines (n 15) G-VII, 3, 5. 96 EPO, Guidelines (n 15) G-VII, 3. 97 EPO, Guidelines (n 15) G-VII, 3. 98 EPO, Guidelines (n 15) G-VII, 3. 89

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

For present purposes, the EPO formulation performs an overlapping role with the CGK inquiry in Singapore and UK law. 99 It does not freeze the skilled person’s toolkit at a pre-AI moment,100 but neither does it attribute whatever the most advanced AI system can do. The question is what means and capacity for routine work and experimentation were normal in the field at the relevant date. 101 Where AI-assisted analysis becomes normal in a technical field, the EPO’s own formulation is capable of accommodating its inclusion within the skilled person’s assumed capabilities. 102 That development could be accommodated without amending Article 56 EPC, provided that the existing problem-solution approach is applied with an accurate understanding of the skilled person’s ordinary technical environment. 103 Comparative Point of Convergence The Singapore and UK structured approaches differ from the EPO problem-solution approach, but the underlying inquiry is similar. Each system requires the decision-maker to identify the skilled person at the relevant priority date before deciding whether the claimed invention was obvious. In Singapore and the United Kingdom, this requires attention to the skilled person’s CGK and ordinary technical capacities. The EPO similarly attributes to the skilled person the means and capacity for routine work and experimentation normal for the field. 104 The convergence between the three systems is functional rather than terminological. In Singapore, the proposed framework operates through the identification of the skilled person and CGK at the opening stage of the Windsurfing inquiry, together with evidence of the resources and competence ordinarily available to the relevant practitioner or team. In the United Kingdom, the same matters inform the construction of the skilled person, while the significance of an attributed AI-assisted route is assessed through the qualitative and fact-sensitive considerations identified in Actavis.

General Tire (n 15) 482–83; EPO, Guidelines (n 15) G-VII, 3. EPO, Guidelines (n 15) G-VII, 3; Thaler (n 14) [56]–[65], [73]; T 1193/23 (n 31) [1.1.1]. The contrast drawn in the text is an application of those materials. 101 EPO, Guidelines (n 15) G-VII, 3. 102 EPO, Guidelines (n 15) G-VII, 3; cf T 0702/20 (n 29) [14]–[19]. 103 Convention on the Grant of European Patents (European Patent Convention) (adopted 5 October 1973, entered into force 7 October 1977) art 56; EPO, Guidelines (n 15) G-VII, 3, 5. 104 First Currency Choice (n 15) [28], [31], [41]–[44]; Mühlbauer (n 15) [20]–[21]; EPO, Guidelines (n 15) G-VII, 3, 5. 100

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

Under the EPO approach, CGK is considered alongside the means and capacity for routine work and experimentation normal in the field. 105 None of these approaches treats AI tools generally as part of CGK or presumes that the skilled person ordinarily works with AI. The existence, public accessibility or scientific prominence of a tool is insufficient. The evidence must establish that the particular tool or function formed part of ordinary technical practice at the priority date. Even where that is shown, attribution establishes only that the relevant research route was available. Whether the claimed result was obvious remains governed by the established test in the relevant jurisdiction. 106 That distinction frames the jurisdiction-specific analysis that follows. It establishes the factual baseline against which each system applies its own test of obviousness. 107 III.

INVENTIVE STEP FRAMEWORKS IN SINGAPORE, THE UNITED KINGDOM, AND THE EPO

Introductory Observation The following jurisdiction-specific analysis demonstrates that the case for an AI-augmented skilled person does not depend on the adoption of a new conception of inventive step. 108 Each system already contains mechanisms for identifying what the skilled person knew, the resources ordinarily available to that person and the work that could routinely be performed at the priority date.109 A.

Singapore

Statutory Framework Section 15 of the Singapore Patents Act provides that an invention involves an inventive step if it is not obvious to a person skilled in the art, having regard to matter forming part of the state of the

105

First Currency Choice (n 15) [28], [31], [41]–[45]; Actavis (n 7) [59]–[73], [88]; EPO, Guidelines (n 15) G-VII, 3, 5. 106 General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) GVII, 3, 5. 107 Actavis (n 7) [59]–[73]; First Currency Choice (n 15) [44]–[51]. 108 First Currency Choice (n 15) [41]–[45]; Actavis (n 7) [59]–[60]; EPO, Guidelines (n 15) G-VII, 3, 5. 109 First Currency Choice (n 15) [35]–[44]; EPO, Guidelines (n 15) G-VII, 3, 5.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

art.110 The provision closely follows section 3 of the UK Patents Act 1977, and that relationship has shaped Singapore doctrine. Singapore courts have consistently treated UK patent authorities as highly persuasive when interpreting inventive step.111 The present issue therefore does not require a separate AI-specific doctrine. It requires the existing skilled-person inquiry to be applied to contemporary technical practice.112 The Skilled Person and Common General Knowledge in Singapore The leading Singapore authority on the structured approach to inventive-step is First Currency Choice. The Court of Appeal treated the Windsurfing framework as a useful means of organising the obviousness inquiry rather than a rigid test, while the IPOS Examination Guidelines note that the later Pozzoli reformulation has not been formally adopted by the Singapore courts. The value of the structure lies in fixing the legal benchmark before the tribunal reaches the ultimate question of obviousness.113 The skilled person is a notional practitioner rather than an actual researcher of exceptional ability. Where the technical problem calls for several forms of expertise, the benchmark may represent a team whose members collectively possess the knowledge and skills ordinarily required for the work.114 This flexibility matters in AI-intensive research, where a conventional research team may include domain specialists and practitioners familiar with data analysis, modelling or established machine-learning tools. Recognising that expertise does not turn the skilled person into an AI system. It simply reflects how the relevant technical work was ordinarily organised at the priority date. The structured inquiry also provides protection against hindsight. Once it is known that an AI system generated, refined or identified the route to the invention, the benchmark may be distorted in either direction. The exclusion of tools that ordinary researchers already used at the priority date may understate what the skilled person could ordinarily achieve, while attributing the capabilities of frontier AI may make the benchmark unrealistically demanding. The skilled person should

110

Patents Act 1994 (2020 Rev Ed), s 15. First Currency Choice (n 15) [35]–[44]; Mühlbauer (n 15) [18]–[21]; Ng-Loy (n 26) paras 30.2.8–30.2.15. 112 Patents Act 1994 (2020 Rev Ed), s 15; Patents Act 1977, s 3; Ng-Loy (n 26) paras 30.2.8–30.2.15. 113 First Currency Choice (n 15) [41]–[45]; IPOS, Examination Guidelines (n 15) paras 4.18–4.24. 114 First Currency Choice (n 15) [28], [31]; Actavis (n 7) [59]–[60]. 111

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

therefore be constructed from evidence of practice at the priority date, rather than adjusted retrospectively to fit the route by which the invention was reached. 115 Mühlbauer later restated the structured inquiry and described the notional addressee as ordinarily skilled, competent and unimaginative. 116 The decision also cautions against treating the exceptional knowledge or ability of an expert witness as representative of the skilled person. 117 Mühlbauer did not, however, establish a separate test for deciding whether a particular tool or research method formed part of CGK. The court expressly observed that the difficulties concerning CGK discussed in First Currency Choice did not arise for determination in that appeal. The status of an AI tool is therefore better addressed through the established principles governing CGK and the wider construction of the skilled person’s ordinary technical environment. 118 Those principles require more than proof that an AI tool existed or was publicly available. Information forms part of CGK only where it was generally known and accepted by the relevant practitioner community as a sound basis for further action. 119 The IPOS Examination Guidelines draw a similar distinction between generally accepted working knowledge and material that is merely capable of being found through a search. 120 In the AI context, CGK therefore concerns whether practitioners knew of the tool, understood its established uses and regarded it as sufficiently reliable for the technical task in question.121 Questions of practical access, supporting infrastructure and the ability to deploy the tool belong instead to the broader inquiry into the skilled person’s ordinary technical means. 122 That assessment may produce different answers across disciplines and even across different uses of the same system.123 A tool may, for example, be routinely relied upon for preliminary screening 115

First Currency Choice (n 15) [44]–[45], [51]; Mühlbauer (n 15) [101]–[102]; Mandel (n 81) 1421–23; Abbott, ‘Everything Is Obvious’ (n 9) 8–11; Schellekens (n 9) 93–97. 116 Mühlbauer (n 15) [20]–[21]. 117 Mühlbauer (n 15) [48]. 118 Mühlbauer (n 15) [20]; General Tire (n 15) 482–83; First Currency Choice (n 15) [38]–[41], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 119 First Currency Choice (n 15) [38]–[41], [50]; General Tire (n 15) 482–83. 120 IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 121 General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 122 First Currency Choice (n 15) [28], [31]; Actavis (n 7) [59]–[60]; EPO, Guidelines (n 15) G-VII, 3. The distinction between CGK and ordinary technical means is the author’s analysis. 123 General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) GVII, 3.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

while remaining insufficiently trusted for final candidate selection or as a substitute for experimental validation. 124 Its prominence in academic discussion or availability online therefore does not, without evidence of ordinary reliance and realistic use, establish that it formed part of the skilled person’s working environment. 125 The inquiry must remain specific to the relevant field, technical purpose and priority date. 126 Towa Corp v ASM Technology further illustrates the evidential discipline required when obviousness is alleged. The party challenging the patent must identify the relevant prior art and explain why the differences between that material and the claimed invention would have been obvious to the skilled addressee. In an AI-assisted case, it is not enough to assert that a machinelearning system could have generated the result. The evidence should show that the relevant route was ordinarily available and that following it would not itself have required inventive judgment. 127 These authorities do not create a separate inventive-step rule for AI-assisted inventions. They support the need for an accurate construction of the skilled person before obviousness is assessed. On the approach advanced here, the court should determine what the practitioner or team knew, which technical resources were realistically available and how those resources were ordinarily used at the priority date. Only then should it ask whether the claimed invention would have been obvious. IPOS Guidance on AI-Related Applications The same broad picture emerges from IPOS examination practice. The Examination Guidelines for Patent Applications at IPOS set out the Windsurfing framework, discuss the later Pozzoli reformulation and describe CGK in terms that closely track the case law. 128 Examiners are directed to consider what would have been generally known and accepted as a sound basis for further action by the bulk of those engaged in the relevant art. 129 Properly understood, that formulation is capable

124

Elton and others (n 3) 828–49; Freilich and Rai (n 11) 924–26; Actavis (n 7) [62]–[73]. General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) GVII, 3. 126 General Tire (n 15) 482–83; First Currency Choice (n 15) [38]–[44], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 127 Towa Corp v ASM Technology Singapore Pte Ltd and another [2016] SGHC 280, [2017] 3 SLR 771, [60]–[64]. 128 First Currency Choice (n 15) [41]–[45]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32, 4.18–4.24. 129 IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 125

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

of extending beyond propositions of abstract technical knowledge to the tools, methods and research practices ordinarily relied upon by practitioners. The 2024 Supplemental Guidance for the Examination of AI-Related Patent Applications is expressly supplementary and non-binding.130 Through a series of hypothetical examples, it explains how AI-related claims are assessed under the existing Examination Guidelines. 131 It also confirms that inventive step, support and complete disclosure remain governed by the requirements applicable to other computer-implemented inventions.132 IPOS has therefore chosen to accommodate AI-related inventions within established patent doctrine rather than create a separate standard of patentability. The Guidance does not, however, resolve the distinct question examined in this article. Although it considers the patentability of inventions involving AI, it does not identify when access to, or use of, a particular AI system should be attributed to the skilled person as part of ordinary technical practice.133 Nor does it provide criteria for determining when a tool moves from being merely accessible to being part of the equipment and working methods ordinarily used in the field. Its significance is therefore primarily institutional: it confirms IPOS’s willingness to apply established patent principles to technological developments, while leaving the common general knowledge inquiry to be worked out through existing doctrine. The framework proposed in this article addresses that unresolved issue by distinguishing between the skilled person’s accepted technical knowledge and the means ordinarily available for applying that knowledge. Commercial accessibility is relevant, but availability alone is insufficient. 134 The more demanding questions are whether the tool had been routinely adopted in the relevant field and whether it could ordinarily be deployed by practitioners at the priority date. These criteria do not displace the Guidance. They provide a structured method for applying the established inventive-step inquiry to a question that the Guidance leaves open.

130

IPOS, Supplemental Guidance (n 27) paras 1–2. ibid paras 1–5. 132 ibid paras 2, 16–17. 133 ibid paras 1–5, 15–17. 134 General Tire (n 15) 482–83; First Currency Choice (n 15) [38]–[44], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 131

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

Taken together, Singapore’s statutory framework, judicial doctrine and administrative practice support an approach that is stable in principle but capable of responding to changes in technical practice. The skilled person remains a notional human practitioner, or team of practitioners, assessed at the priority date and without hindsight.135 That human-centred character does not require the inquiry to disregard the tools through which ordinary technical work is performed. Singapore’s skilled-person inquiry is sufficiently flexible to recognise tools and methods that have become embedded in ordinary technical practice, while preserving the distinction reflected in General Tire and First Currency Choice between material that is merely available and material that is generally known, accepted and relied upon by practitioners. 136 Singapore’s close alignment with UK patent jurisprudence, together with IPOS’s use of supplementary AI guidance, 137 places it in a strong institutional position to develop this approach incrementally through examination practice and, where necessary, judicial reasoning. 138 B.

United Kingdom

Statutory and Doctrinal Framework Section 3 of the Patents Act 1977 provides that an invention shall be taken to involve an inventive step if it is not obvious to a person skilled in the art, having regard to matter forming part of the state of the art.139 UK law has developed a substantial body of authority on the skilled person, common general knowledge and the assessment of obviousness. That jurisprudence is especially relevant to Singapore because Singapore courts regularly draw on UK patent decisions when applying section 15 of the Patents Act. 140 Developments in UK law are therefore likely to inform the arguments available when the effect of routine AI use on the skilled-person inquiry comes before a Singapore court.

135

First Currency Choice (n 15) [28], [31], [41]–[45]; Mühlbauer (n 15) [20]–[21], [48]. General Tire (n 15) 482–83; First Currency Choice (n 15) [38]–[41], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32. 137 First Currency Choice (n 15) [35]–[44]; Mühlbauer (n 15) [18]–[21]; IPOS, Supplemental Guidance (n 27) paras 1–5. 138 IPOS, Examination Guidelines (n 15) paras 2.24–2.32, 4.18–4.24; IPOS, Supplemental Guidance (n 27) paras 1– 5, 15–17. 139 Patents Act 1977 (UK) s 3. 140 First Currency Choice (n 15) [35]–[44]; Mühlbauer (n 15) [18]–[21]; Ng-Loy (n 26) paras 30.2.8–30.2.15. 136

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

The Windsurfing/Pozzoli Framework in Practice In Actavis, Lord Hodge emphasised that obviousness is a qualitative, fact-sensitive assessment. 141 The court must reconstruct what the skilled person or team would have done at the priority date, having regard to the relevant common general knowledge, the established practices of the field and the resources ordinarily available. These matters are established through evidence rather than assertion. In an AI-assisted case, expert evidence may therefore be relevant to whether practitioners were aware of a particular tool, regarded it as a sound basis for the relevant technical task, could realistically access it and possessed the ordinary competence needed to deploy it. Those questions should not be collapsed into a single inquiry. Awareness of the tool and acceptance of its established uses bear principally on common general knowledge. Access to the necessary software, data and computing infrastructure, together with the competence required to use them, form part of the wider technical environment attributed to the skilled person. Evidence that an AI system was publicly available would not, by itself, establish either proposition. Actavis concerned a patent claiming a low daily dose of tadalafil for treating erectile dysfunction.142 Although the efficacy of the 5 mg dose was described as “rather surprising”, the Supreme Court concluded that the claimed dosage regime was obvious because the skilled team would have arrived at it through the ordinary programme of dose-ranging studies used in pharmaceutical development. 143 The unexpected nature of the result was relevant, but it did not outweigh the routine character of the research programme that led to it. The significance of Actavis is not that every result obtained through a familiar research method is necessarily obvious. Lord Hodge identified a range of considerations, including whether the research was obvious to try, its routine nature, its cost and burden, the value judgments required, the existence of alternative pathways, the skilled person’s motivation, the prospects of success and whether the eventual result was unexpected. 144 The weight attached to each consideration depends on the technical and evidential context.

141

Actavis (n 7) [59]–[60], [88]. Actavis (n 7) [57]–[58]. 143 Actavis (n 7) [64]–[66], [71]–[72]. 144 Actavis (n 7) [62]–[73], [88]. 142

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

The same discipline should apply to AI-assisted research. Attributing a routinely used AI tool to the skilled person may establish that a particular method of search, prediction or optimisation formed part of the ordinary technical environment. It does not establish, without more, that the skilled team would have framed the problem in the same manner, selected the same inputs, relied on the same outputs or reasonably expected the claimed technical result. A routine AI-assisted process may produce an unexpected result, but surprise is only one consideration within the overall assessment. The availability of the route and the obviousness of the claimed destination therefore remain analytically distinct. 145 PHOSITA and CGK in UK Law The UK authorities continue to treat PHOSITA as a notional human practitioner of ordinary skill.146 The familiar description of the normally skilled but unimaginative addressee remains embedded in the Windsurfing/Pozzoli framework. It excludes inventive ingenuity, but does not reduce the skilled person to a passive or poorly equipped worker. The skilled person is competent in the relevant field, possesses ordinary technical judgment and can undertake routine experimentation. Where the work ordinarily requires several forms of expertise, the benchmark may comprise a team rather than an individual. 147 CGK remains anchored in General Tire. It does not encompass every item that could be discovered in the literature. The relevant material must be generally known and generally regarded as a sound basis for further action by the bulk of those engaged in the art. 148 Later authorities have preserved this practical orientation. CGK may include material that practitioners know exists and would consult as a matter of course, but not obscure publications or specialised information that would require a deliberate search to uncover.149 That distinction is important for AI-assisted research. Public or commercial availability alone does not make an AI tool part of CGK. Evidence would still be required that practitioners in the relevant field were generally aware of the tool and accepted its established functions as a reliable basis for

145

Actavis (n 7) [62]–[73], [88]. Windsurfing (n 5) 73–74; Pozzoli (n 5) [23]; Mühlbauer (n 15) [20]–[21]. 147 Actavis (n 7) [59]–[60]. 148 General Tire (n 15) 482–83. 149 Halliburton (n 62) [13]. 146

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

further work. Access to the tool, the infrastructure required to operate it and competence in its routine use may also form part of the wider technical environment attributed to the skilled person. Existing doctrine therefore permits a middle position. The law need not choose between a purely analogue skilled person and one whose reasoning is replaced by AI. Thaler and the Anthropocentric Architecture In Thaler, the UK Supreme Court held that an inventor within the meaning of the Patents Act 1977 must be a natural person. 150 DABUS could not qualify because it was a machine without legal personality. Lord Kitchin’s reasoning rested on the structure and language of sections 7 and 13, including the definition of the inventor as the “actual deviser” of the invention. 151 The decision concerned both inventorship and the statutory route by which an applicant becomes entitled to the grant of a patent. It did not establish that the inventor must personally own the resulting rights. The Act permits a patent to be granted to an employer, successor or other person claiming through the inventor, but the statutory chain must begin with an inventor who is a natural person.152 The skilled-person inquiry performs a different function. PHOSITA does not devise, author or own the invention. It is an objective legal benchmark used to determine whether the claimed invention would have been obvious. Attributing to that human benchmark the tools, resources and working methods ordinarily used by real practitioners does not recognise the tool itself as an inventor. Thaler therefore does not foreclose the argument advanced here. Both approaches remain humancentred, but they answer different questions: Thaler identifies who may qualify as the inventor, while the present inquiry determines the knowledge, competence and resources attributed to the person assessing obviousness. Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks In Emotional Perception, the UK Supreme Court reconsidered the treatment of computerimplemented inventions under the excluded-subject-matter provisions. It held that step 1 of the

150

Thaler (n 14) [56]–[65], [73]. Thaler (n 14) [28]–[32], [59]–[63], [73]–[79]; Patents Act 1977, ss 7(2), 13(2). 152 Patents Act 1977, s 7(2)(b)–(c); Thaler (n 14) [57]–[63], [73]–[79]. 151

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

Aerotel approach, requiring the claim to be construed, remained unobjectionable, but that steps 2– 4 could no longer be maintained in light of the interpretation of article 52 of the European Patent Convention adopted in G 1/19.153 Under the revised approach, a claim involving technical means may qualify as an invention for the purposes of article 52. Where a claim contains both technical and non-technical features, an intermediate exercise must then identify which features contribute to the technical character of the invention. Features that make no such contribution are excluded before novelty and inventive step are assessed.154 The Court did not, however, require UK courts to adopt the EPO’s problemsolution approach. It expressly confirmed that the Pozzoli method remains a legitimate means of assessing inventive step under UK law. 155 The decision does not resolve the question considered in this article. It concerns whether a computer-implemented claim qualifies as an invention and how non-technical features should be treated before novelty and inventive step are examined. It does not determine how the skilled person should be calibrated where AI tools have become part of ordinary research practice. The Court also recognised that its conclusions might have repercussions for the methodologies used to assess novelty and inventive step. It had not received submissions directed to that issue and therefore left it to a future case. 156 Nor did it determine precisely how the new intermediate exercise should be applied to the claims before it. The Court considered it unwise to create binding precedent on that question without fuller argument, particularly given the importance of the issue to computer-implemented inventions and AI. 157 Its significance for the present argument is therefore indirect. More computer-implemented claims involving technical means may proceed beyond the article 52 inquiry and reach the ordinary requirements of novelty and inventive step. Once they do, the court must still identify the skilled person, determine the relevant CGK and decide which technical features are capable of supporting

153

Emotional Perception AI Limited v Comptroller General of Patents, Designs and Trade Marks [2026] UKSC 3, [54]–[65]; G 1/19 Pedestrian Simulation [2021] OJ EPO A77, [28], [37]–[39]. 154 Emotional Perception AI Limited (n 153) [65]–[67], [100]–[107]. 155 Emotional Perception AI Limited (n 153) [54], [65]–[67], [116]. 156 Emotional Perception AI Limited (n 153) [67], [116]. 157 Emotional Perception AI Limited (n 153) [99]–[118], especially [115]–[118].

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

inventive step. The framework developed here addresses that downstream inquiry. This is an inference from the consequences of Emotional Perception, rather than an approach expressly endorsed by the Supreme Court. Taken together, the UK authorities support a human-centred but technologically responsive approach. Thaler preserves the requirement of human inventorship. General Tire, Pozzoli and Actavis allow the obviousness inquiry to reflect the knowledge, methods and resources ordinarily found in the relevant technical field. Emotional Perception directs AI-related and other computerimplemented claims into a revised patentability framework, but leaves open how the skilled person should be calibrated when those claims reach the inventive-step stage. C.

European Patent Office

Statutory Framework and the Skilled Person Under the EPO Guidelines Article 56 of the European Patent Convention provides that an invention involves an inventive step if, having regard to the state of the art, it is not obvious to a person skilled in the art. 158 The EPO Guidelines describe that person as a practitioner in the relevant field with average knowledge and ability, who is aware of the common general knowledge existing at the relevant date. 159 The skilled person is also presumed to have access to the state of the art and to possess “the means and capacity for routine work and experimentation which are normal for the field of technology in question”.160 That formulation overlaps with the Singapore and UK approaches, although the concepts should not be treated as identical. The EPO’s presumption of access to the state of the art is broader than common general knowledge. For present purposes, the more relevant feature is the attribution of the means and capacity for routine work normal in the field. It allows the resources attributed to the skilled person to reflect changes in ordinary technical practice without altering the human character of the benchmark.

158

European Patent Convention, art 56. EPO, Guidelines (n 15) G-VII, 3. 160 EPO, Guidelines (n 15) G-VII, 3. 159

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

Board of Appeal Decisions on AI and Machine Learning The EPO Boards of Appeal have developed a substantial body of decisions concerning whether machine-learning features contribute to the technical character of an invention. These decisions do not directly define the tools and resources ordinarily available to the skilled person. Their immediate concern is whether the claimed computational features solve a technical problem or produce a credible technical effect. In T 0702/20 (Sparsely Connected Neural Network/Mitsubishi), the Board considered a neuralnetwork architecture in which connections were determined by the parity-check matrix of an errorcorrecting code.161 It characterised a neural network, as such, as a class of mathematical functions. Such a feature could be considered in the assessment of inventive step only where it contributed to the solution of a technical problem, for example through training with specific data for a specific technical task.162 The Board also rejected the applicant’s attempt to equate the neural network with a human brain. Although a neural network’s structure may be inspired by biological systems, there was no evidence that it functioned in the same manner as the human brain. Its behaviour remained determined by its structure, training data and training procedure. 163 This conclusion prevents technological complexity or apparent unpredictability from being treated, without more, as evidence of human-like cognition or technical contribution. This line of reasoning is reinforced in T 1425/21 (Distilled Machine Learning Models/Google), which concerned the distillation of large machine-learning models into a smaller model with reduced memory requirements.164 The Board accepted that the distilled model required less memory, but held that reduced storage or computational requirements were insufficient by themselves to establish a technical effect. The performance of the smaller model also had to be considered.165

161

T 0702/20 (n 29) [1]–[1.2]. T 0702/20 (n 29) catchword, [12]–[14], [19]. 163 T 0702/20 (n 29) [15]–[16.2]. 164 T 1425/21 Distilled Machine Learning Models/Google (EPO Technical Board of Appeal, 7 February 2024) [1]– [2.1]. 165 T 1425/21 (n 164) [15]–[18]. 162

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

The alleged effect was not credible across the full scope of the claim. The application did not establish that the smaller models would retain comparable accuracy, explain how the relevant temperature parameters should be selected, or provide results showing which distilled models could replace which larger models. 166 The decision therefore concerns the credibility and scope of the alleged technical effect, rather than establishing that model distillation is necessarily routine or inventive. The same approach was applied in T 1952/21 (Reinforcement Learning/Bosch), concerning a reinforcement-learning system for controlling an industrial process. The Board similarly rejected the proposition that reinforcement learning possesses technical character merely because it may be used in technical applications. 167 An agent and its environment could be entirely virtual, including within a simulation or video game. Reinforcement learning in general therefore did not imply a technical context. A trained machine-learning model could contribute to inventive step where it was used to solve a technical problem, such as through training with specific data for a specific technical task, but that connection had to be present in the claim. 168 T 1998/22 (Wide and Deep Machine Learning Models/Google) extended the same reasoning to a system combining wide and deep machine-learning models.169 The claimed method processed abstract data through mathematical computational models. Joint training through backpropagation did not confer technical character, particularly where the models had not been trained to perform a specified technical function. 170 These decisions do not determine whether a machine-learning tool formed part of the skilled person’s common general knowledge or ordinary technical environment. They establish a different proposition. AI terminology, mathematical complexity and the use of a trained model do not themselves supply a technical contribution. The claimed features must contribute to the solution of a technical problem or produce a credible technical effect across the relevant scope of the claim.

166

T 1425/21 (n 164) [19]–[20.2]. T 1952/21 (n 29) [1]–[4]. 168 T 1952/21 (n 29) [22]–[24.1]; T 0702/20 (n 29) catchword. 169 T 1998/22 Wide and Deep Machine Learning Models/Google (EPO Technical Board of Appeal, 20 December 2024) [1]–[3]. 170 T 1998/22 (n 169) [11]–[11.2]. 167

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

Their relevance to the present argument is therefore limited but important. They demonstrate why two inquiries must remain separate. One asks whether a particular AI tool or method was ordinarily available to the skilled person at the priority date. The other asks whether the AI-related features of the claimed invention made a technical contribution. Routine access to a tool may bear on the first inquiry, but it does not answer the second. AI-Generated Evidence and T 1193/23 T 1193/23 concerned a different use of AI. During opposition appeal proceedings, the patent proprietor relied on answers generated by ChatGPT to support its interpretation of expressions used in a patent claim. 171 The answers had been presented orally and were not placed in full on the record. The Board nevertheless addressed their evidential significance and stated that a ChatGPT answer was irrelevant in itself because claim interpretation depended on the understanding of the skilled person.172 The increasing prevalence of large-language-model chatbots did not justify assuming that an answer generated from training data unknown to the user, and sensitive to the context and wording of the prompt, correctly represented the understanding of the skilled person in the relevant technical field at the relevant date. That understanding could instead be established through appropriate technical literature. 173 The decision imposes a narrow but significant evidential limit. AI-generated text cannot, merely because it is produced by a widely used system, substitute for evidence of the skilled person’s understanding. It would be too broad, however, to treat T 1193/23 as a general ruling that AI tools can never form part of the skilled person’s technical environment. The Board addressed the reliability of chatbot output as evidence of claim meaning. It did not decide whether a particular AI system might be attributed to the skilled person as a research tool where routine adoption, access and ordinary competence were independently established. Read with the Guidelines, the decisions support a restrained, human-centred position. The skilled person may possess the means and capacity for routine work deemed normal in the field. In 171

T 1193/23 (n 31) [1.1.1]. T 1193/23 (n 31) [1.1.1]. 173 T 1193/23 (n 31) catchword, [1.1.1]. 172

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

principle, that formulation is capable of encompassing an AI tool where its use had genuinely become part of ordinary technical practice. The Board decisions do not themselves establish when that threshold is crossed. They do confirm that attribution of the tool would not make its output determinative of obviousness, technical effect or the skilled person’s understanding. The legal assessment remains centred on the knowledge, judgment and ordinary capacities of the human practitioner or team. IV.

AI-ASSISTED INNOVATION AND EMERGING DOCTRINAL TENSIONS A.

A Typology of AI Involvement in the Inventive Process

Before turning to the doctrinal tensions created by AI-assisted innovation, it is necessary to distinguish between different levels of AI involvement in the inventive process. Not every use of AI raises the same legal problem, and much of the confusion in current debates stems from treating materially different forms of machine involvement as though they were interchangeable. 174 For present purposes, three levels of AI involvement can be identified. Level 1 is AI as a Research Tool. At this level, a human researcher identifies the technical problem, selects the AI system to be used and configures that system to generate, screen or optimise candidate solutions. The researcher then exercises independent technical judgment in interpreting, validating and refining the outputs. The human contribution remains central: AI accelerates or expands the research process, but does not displace the researcher as the principal intellectual actor. This describes much of the contemporary use of AI in protein-structure prediction, retrosynthesis planning, engineering-design optimisation and machine-learning screening in molecular and materials science. 175 Level 2 is AI as a Material Contributor. Here, AI contributes substantially to the generation of the inventive concept itself, while the researcher continues to frame the problem, select or configure the model, assess its outputs and determine which results to pursue. The human role remains real, but the core technical idea emerges less from conventional human reasoning than from AI-

174

Abbott, ‘Everything Is Obvious’ (n 9) 23–28; Schellekens (n 9) 91–93. Jumper and others (n 1) 583–89; Marwin HS Segler, Mike Preuss and Mark P Waller, ‘Planning Chemical Syntheses with Deep Neural Networks and Symbolic AI’ (2018) 555 Nature 604, 604–10; Butler and others (n 3) 547–55; Mirhoseini and others (n 3) 207–12. 175

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

mediated exploration of the solution space. The invention is no longer easily characterised as either human-devised with incidental computational assistance or autonomously generated by AI. It occupies an intermediate space in which human judgment and computational search are both materially involved.176 Singapore inventorship doctrine provides a useful, but limited, point of comparison. In Dien Ghin, the High Court treated the inventor as the actual deviser of the invention and focused on who had formulated the inventive concept, even where others subsequently worked out the details required to put it into effect.177 That inquiry concerns the attribution of inventorship. It does not determine the objective benchmark against which inventive step is assessed. The decision may therefore assist in cases near the boundary between Levels 1 and 2, where the question is whether a human researcher made a sufficient contribution to the inventive concept despite substantial AI involvement. It should not, however, be used to define the person skilled in the art, the content of common general knowledge or the ordinary technical resources relevant to obviousness. Level 3 is Autonomous AI Invention. At this level, an AI system independently identifies the technical problem and generates the solution without meaningful human intellectual direction. This is the DABUS-style scenario addressed in the Thaler litigation, which has occupied a prominent place in debates concerning AI and inventorship.178 For present purposes, the more immediate doctrinal pressure lies at Level 1 rather than Level 3. AI-assisted methods are already used across several research-intensive fields and may affect the factual baseline against which inventive step is assessed. 179 Once ordinary researchers in a field routinely use particular AI tools as part of standard research practice, the practical question shifts

176

Abbott, ‘Everything Is Obvious’ (n 9) 23–31; Shemtov (n 3) 7–9; Freilich and Rai (n 11) 924–26. Dien Ghin Electronic (S) Pte Ltd v Khek Tai Ting (trading as Soon Heng Digitax) [2011] SGHC 36, [2011] 3 SLR 227, [13]; Cicada Cube Pte Ltd v National University Hospital (Singapore) Pte Ltd and another appeal [2018] SGCA 52, [2018] 2 SLR 940, [61]–[62], [86]. 178 Thaler (n 14) [1]–[3], [56]–[65], [73]. 179 Shemtov (n 3) 7–9; Jumper and others (n 1) 583–89; Segler, Preuss and Waller (n 175) 604–10; Butler and others (n 3) 547–55; Freilich and Rai (n 11) 924–26. 177

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

from whether AI can be an inventor to whether obviousness can properly be assessed through a skilled person imagined as working without those tools. The boundary between Levels 1 and 2 nevertheless remains important. Freilich and Rai’s study of AI-derived drug patents suggests that some AI-native firms may be doing more than using AI as a screening accelerant. In some cases, an AI system may generate the candidate molecule, model aspects of its predicted efficacy and shape the route towards the claimed invention, while the human role is concentrated in problem framing, selection and validation. 180 That does not amount to full autonomy, but neither is it straightforwardly equivalent to the use of a neutral research instrument. The framework developed in this article is most persuasive where AI forms part of the ordinary process of discovery. It separates the knowledge attributed to the skilled person from access to the relevant tool and the competence needed to use it in practice. Its reach is more limited where AI itself appears to have generated the inventive concept, but that boundary is important because it defines the precise problem the framework is intended to address. B.

The Low-Threshold Risk: Over-Patenting Routine AI Outputs

The first doctrinal risk arises where knowledge of, practical access to and ordinary competence in using AI tools are excluded from the resources attributed to the skilled person. In that situation, the inventive-step threshold is calibrated against a notional practitioner who does not use tools that ordinary researchers in the relevant field routinely deploy. The benchmark may then become artificially weak. Where practically accessible AI systems are routinely used to identify candidate compounds, predict protein structures, optimise materials properties or explore engineering-design spaces, patents over the ordinary products of those processes may reward effective tool use rather than inventive contribution.181 Patent protection may then favour those best placed to deploy standard

180

Freilich and Rai (n 11) 924–26. Jumper and others (n 1) 583–89; Butler and others (n 3) 547–55; Mirhoseini and others (n 3) 207–12; Freilich and Rai (n 11) 924–26. 181

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

computational tools quickly and strategically, even where the claimed advance does not go beyond ordinary technical work. The concern is particularly acute in research-intensive sectors such as pharmaceuticals and biotechnology. AI may alter not only the speed of research, but also the baseline of what counts as routine work. A task that once required prolonged and costly experimental effort may become substantially more manageable through the ordinary deployment of a known model, standard inputs and domain-specific judgment.182 If the skilled-person benchmark is not adjusted to reflect that change, the patentability threshold may fall without any express alteration of doctrine or policy. Freilich and Rai’s empirical study makes the concern more concrete. They found that AI-native pharmaceutical companies had filed compound patents based on AI-identified molecular candidates with materially less evidence of real-world testing than traditional pharmaceutical developers.183 That finding does not establish that the patents were invalid or that the claimed compounds were necessarily obvious. It indicates, however, that AI-assisted methods may permit commercially significant candidates to be identified and placed within patent claims earlier and at lower cost than conventional research methods. Against a legal benchmark that excludes tools ordinarily used in the field, those outputs may appear inventive even where they are substantially the products of routine algorithmic search. The patent system would then risk protecting the possession and effective use of standard computational capacity rather than the presence of an inventive technical contribution. The potential institutional consequences resemble the anticommons dynamic identified by Heller and Eisenberg.184 An artificially low threshold may increase the risk of over-patenting, particularly in cumulative research sectors. The result could be a denser network of exclusionary rights over subject matter that ordinarily skilled researchers, equipped with standard AI tools, might have reached without inventive ingenuity. Downstream research costs may rise, follow-on innovation

182

Abbott, ‘Everything Is Obvious’ (n 9) 23–31; Schellekens (n 9) 91–98; Butler and others (n 3) 547–55. Freilich and Rai (n 11) 924–25. 184 Heller and Eisenberg (n 10) 698–701. 183

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

may become more difficult and the patent system may operate less effectively as a filter for genuinely inventive contributions. C.

The High-Threshold Risk: Collapsing Inventive Step into AI Capability

The converse danger is that too much is attributed to the skilled person. A benchmark endowed with the full problem-solving capacity of the most advanced AI systems available at the priority date would bear little resemblance to the ordinary practitioner on whom inventive-step doctrine is based. Technical solutions that remain costly, uncertain or difficult for ordinary research teams might nevertheless be treated as obvious because an advanced model could, at least in principle, identify them. Such an approach could suppress patent protection just as an artificially weak benchmark could expand it. The inquiry would begin to resemble hindsight exercised with idealised computational capabilities, rather than an assessment of what an ordinarily skilled practitioner would have done at the priority date. The effect would be particularly pronounced in fields where AI systems are developing quickly and access to frontier capabilities remains uneven. This would also depart from the established conception of the skilled person. PHOSITA is not a figure of maximal technical power, but a practitioner of ordinary competence who possesses the knowledge, judgment and capacity for routine work normally found in the relevant field.185 The absence of inventive imagination does not imply the possession of every capability that technology can provide. Treating the benchmark as though it could draw without restriction on the most advanced AI systems would replace ordinary technical competence with exceptional machine capability. That distinction matters in pharmaceuticals and other fields in which identifying a promising candidate is only one stage of a longer research process. Even where AI substantially narrows the search space, researchers may still have to formulate the problem, curate and structure the inputs, select and configure the model, evaluate the reliability of its outputs and undertake experimental

185

First Currency Choice (n 15) [28], [31], [41]–[44]; Mühlbauer (n 15) [20]–[21]; EPO, Guidelines (n 15) G-VII,

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

validation.186 Further development may involve considerable cost, uncertainty and technical risk. An invention should not become obvious merely because AI contributed to its discovery where the claimed result still depended on substantial human judgment or non-routine experimental work. The same point creates a tension, although not a formal inconsistency, with the human-centred approach to inventorship in Thaler. The UK Supreme Court held that an inventor recognised under the Patents Act 1977 must be a natural person. 187 Inventorship and obviousness plainly perform different functions, but it would nevertheless be difficult to justify a system that requires a human inventor while judging that person’s contribution against capabilities beyond those ordinarily available to any human practitioner or team. The problem lies in the resulting asymmetry: human inventorship would be retained, while the ordinary-skill benchmark would be displaced by exceptional non-human capacity. D.

The Central Tension: Available Routes and Claimed Results

The central distinction is between the availability of an AI-assisted research route and the obviousness of the result reached through it. As AI tools become integrated into standard research practice, some forms of computational search, prediction and optimisation may themselves become routine. A researcher might ordinarily use AlphaFold to predict protein structures, an AI synthesis planner to identify possible reaction pathways or a generative-design platform to optimise technical parameters. 188 Whether such a tool should be attributed to the skilled person depends on its position within the relevant field at the priority date. Practical accessibility is necessary, but is not sufficient by itself. The evidence must also show that the tool was routinely adopted, accepted as a reliable basis for the relevant task and capable of being deployed through ordinary professional competence. These considerations do not require every aspect of the tool to be classified as CGK. Awareness of the tool and acceptance of its established uses may form part of the skilled person’s knowledge, while

186

Freilich and Rai (n 11) 924–26; Elton and others (n 3) 828–49; Butler and others (n 3) 547–55; Shemtov (n 3) 7–

9. 187 188

Thaler (n 14) [56]–[65], [73]. Jumper and others (n 1) 583–89; Segler, Preuss and Waller (n 175) 604–10; Butler and others (n 3) 547–55.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

access to the necessary software, data and infrastructure, together with competence in operating them, belongs to the wider technical environment in which that knowledge is applied. Establishing that an AI-assisted route was available does not determine whether the claimed invention was obvious. The skilled person may have known of and been capable of using the relevant tool without necessarily selecting it for the problem at hand, framing the problem in the same way, choosing the same inputs or placing confidence in the resulting output. These matters remain part of the ordinary, fact-sensitive assessment of what the skilled person would have done. A routinely available method does not make every result produced through it obvious. The skilled person may have had access to the relevant AI-assisted route, yet the claimed invention may still have depended on an unexpected technical effect, a non-routine choice or judgment extending beyond the ordinary use of the tool. Conversely, the fact that the precise output could not have been predicted in advance does not, by itself, establish an inventive step. The court must consider the research programme as a whole, including whether the skilled person would have pursued that route, the alternatives available, the cost and burden involved, the technical decisions required, the prospects of success and the significance of any unexpected result. 189 The ultimate question is whether the claimed invention would have been obvious to the skilled person at the priority date, taking into account the AI-assisted methods ordinarily available in the field. The assessment must remain specific to the particular tool, task, technical field and priority date. Evidence that practitioners routinely relied on one system for a defined purpose does not establish that its other functions, or AI systems more generally, formed part of ordinary technical practice, still less that adoption in one discipline can be transferred without inquiry to another. The court or patent office should therefore consider how the tool was actually used in the field, including its cost and practical availability, licensing restrictions, data and computing requirements, the training needed to operate it and its reliability for the task in question. 190 In practice, these considerations help distinguish inventions that emerge as foreseeable products of standard AI-assisted workflows from those that may remain non-obvious despite the availability

189

Actavis (n 7) [62]–[73], [88]. General Tire (n 15) 482–83; First Currency Choice (n 15) [38]–[44], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 190

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

of AI. The latter may depend on unexpected technical properties, unusual problem framing, nonstandard inputs, substantial experimental validation or an integration of human and computational work that falls outside ordinary practice. 191 This is not intended to create separate legal categories, but to identify factual features that may guide courts and patent offices when applying the established inventive-step inquiry. The proposed approach avoids both extremes by calibrating the skilled person to ordinary technical practice at the priority date. It neither disregards tools that practitioners routinely used, which might allow ordinary computational outputs to appear inventive, nor attributes frontier machine capabilities that remained beyond their practical reach, which might deny protection to work that still required substantial technical judgment. 192 The knowledge, access and practical competence ordinarily possessed by practitioners determine which research routes were available; whether the claimed result was obvious through those routes remains a separate, fact-sensitive inquiry.193 V.

THE AI-AUGMENTED SKILLED PERSON A.

Three Possible Responses

The effect of AI-assisted research on the inventive-step inquiry may be approached in three ways. The first is to retain a skilled person whose assumed methods and resources do not include AI. Although this position may appear doctrinally conservative, it ceases to be neutral once practitioners in the relevant field routinely rely on AI for screening, prediction, optimisation or other technical tasks. 194 A benchmark constructed without those tools would no longer reflect the conditions under which ordinary research is undertaken. The products of established AI-assisted workflows might then appear inventive, not because they involve a contribution beyond ordinary skill, but because the legal benchmark has been deprived of resources already used in practice. The second approach lies at the opposite extreme. It would attribute to the skilled person the full problem-solving capabilities of frontier AI systems available at the priority date. Such a

191

Actavis (n 7) [62]–[73]; Freilich and Rai (n 11) 924–26; Butler and others (n 3) 547–55. Abbott, ‘Everything Is Obvious’ (n 9) 8–11, 23–31; Schellekens (n 9) 91–98. 193 General Tire (n 15) 482–83; First Currency Choice (n 15) [44]–[51]; Actavis (n 7) [62]–[73], [88]; EPO, Guidelines (n 15) G-VII, 3. 194 Abbott, ‘Everything Is Obvious’ (n 9) 6–9, 35–39; Schellekens (n 9) 92–94, 97. 192

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

benchmark would be detached from ordinary technical practice wherever the relevant system remained inaccessible to most practitioners, required unusual expertise or depended on infrastructure available only to exceptionally well-resourced organisations. A solution that remained costly, uncertain or technically difficult for the ordinary research community might nevertheless be treated as obvious because an advanced system could, in principle, identify it. 195 Both approaches distort the ordinary-skill foundation of inventive step. The first may set the threshold too low by disregarding methods that have become part of standard research practice. The second may set it too high by importing capabilities that ordinary practitioners could not realistically exercise. Neither reflects the established conception of the skilled person as a practitioner of ordinary, rather than maximal, technical competence. 196 The preferable position lies between these extremes. The skilled person should remain a notional human practitioner, or team of practitioners, while the technical environment attributed to that benchmark remains capable of reflecting developments in ordinary research practice. An AI tool should affect that environment only where the evidence establishes that practitioners in the relevant field knew of and accepted the tool for the task in issue, could obtain meaningful access to it, and possessed the competence and resources required for its ordinary deployment. 197 The AI system remains an instrument within the skilled person’s working environment rather than an independent legal actor. The court or patent office must still determine whether the claimed invention would have been obvious to the skilled person using the knowledge, methods and resources that had genuinely become ordinary at the priority date. Establishing that an AI-assisted route was available does not establish, without more, that the claimed result was obvious. 198 B.

The Evidential Framework

The existence or technical significance of an AI system is not sufficient to alter the inventive-step baseline. It is also imprecise to describe software itself, without qualification, as forming part of 195

Abbott, ‘Everything Is Obvious’ (n 9) 6–9, 35–39; Schellekens (n 9) 93–97. First Currency Choice (n 15) [28], [31], [41]–[45]; Mühlbauer (n 15) [20]–[21]; EPO, Guidelines (n 15) G-VII, 3; Abbott, ‘Everything Is Obvious’ (n 9) 6–9; Schellekens (n 9) 93–97. 197 General Tire (n 15) 482–83; First Currency Choice (n 15) [28], [31], [38]–[44], [50]; Actavis (n 7) [59]–[60]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 198 First Currency Choice (n 15) [44]–[51]; Actavis (n 7) [62]–[73], [88]; EPO, Guidelines (n 15) G-VII, 5.3. 196

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

common general knowledge. The analysis should instead distinguish the skilled person’s knowledge of the tool and its accepted uses from practical access to the system and the competence required to employ it for the relevant task.199 Three related considerations structure that inquiry. The first concerns whether the tool was realistically accessible to the relevant practitioner community at the priority date. The second concerns whether practitioners routinely used the tool and regarded it as a reliable basis for the particular technical work in question. The third concerns whether an ordinary practitioner or team could deploy it using the expertise, data and infrastructure normally available in the field. These considerations do not introduce additional statutory requirements or constitute a separate test of patentability. They identify the evidence required to construct the skilled person accurately in an AI-intensive technical setting. 200 Practical accessibility is not established merely because the tool existed or had been publicly disclosed. The relevant practitioner community must have been able to obtain or use it at the priority date, whether through a commercial licence, an open-source release, a publicly hosted platform or an ordinary institutional arrangement. 201 Commercial availability is therefore too narrow a formulation, since a tool may be ordinarily accessible without being sold on the open market. Nominal availability may nevertheless conceal substantial barriers to meaningful use. A system should not be attributed to the skilled person where effective deployment depended on proprietary datasets, unusually expensive computing resources, restrictive licensing arrangements or facilities available only to a small number of well-resourced organisations.202 The inquiry is directed to realistic access within the relevant field rather than technological possibility in the abstract. Accessibility must also be assessed by reference to the particular version and function of the system. The public release of one model or interface does not establish access to later or more

199

General Tire (n 15) 482–83; First Currency Choice (n 15) [28], [31], [38]–[44], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 200 General Tire (n 15) 482–83; First Currency Choice (n 15) [28], [31], [38]–[44], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 201 Schellekens (n 9) 93–96; EPO, Guidelines (n 15) G-VII, 3. 202 Schellekens (n 9) 94–96; Abbott, ‘Everything Is Obvious’ (n 9) 36–39; EPO, Guidelines (n 15) G-VII, 3.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

capable versions. Nor does access to a limited public service necessarily establish access to the underlying model, training process or computational infrastructure. The evidence must therefore identify what was available, to whom and on what terms at the relevant priority date. Access alone does not establish that a tool formed part of ordinary technical practice. A system may be widely publicised, frequently discussed or experimentally tested without being regarded as a sound basis for further technical work. The evidence must address both the extent of adoption and whether practitioners accepted the system as sufficiently reliable for the purpose for which attribution is asserted. 203 That purpose must be defined with precision. Routine use for preliminary screening does not establish that practitioners would rely on the same system to select a final candidate, dispense with experimental testing or support a different technical conclusion.204 A tool may be accepted for exploratory work while remaining unsuitable for decisions carrying greater technical or commercial consequences. Attribution must therefore remain both tool-specific and purposespecific. Accepted reliability does not require certainty or universal confidence. Technical tools are often used despite recognised limitations. The relevant issue is whether practitioners regarded the system as sufficiently dependable to form part of the ordinary basis for further action in the particular context. Isolated use, publicity or general scientific interest is insufficient where ordinary reliance has not followed. The third consideration concerns ordinary deployability. A system should not be attributed to the skilled person where its operation depended on bespoke model development, unusual machinelearning expertise, substantial architectural modification or infrastructure beyond that ordinarily available in the field. 205 The skilled person need not be confined to a single disciplinary background. Patent law recognises that the appropriate benchmark may be a team where the technical task ordinarily calls for several

203

General Tire (n 15) 482–83; IPOS, Examination Guidelines (n 15) paras 2.24–2.32. Elton and others (n 3) 828–49; Freilich and Rai (n 11) 924–26; Actavis (n 7) [62]–[73]. 205 Schellekens (n 9) 93–96; Mühlbauer (n 15) [48]; EPO, Guidelines (n 15) G-VII, 3. 204

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

forms of expertise. 206 The notional team may therefore include a data scientist, computational chemist, structural biologist or other specialist where that combination reflects ordinary practice. The inquiry should not, however, assume capabilities that remained exceptional even among multidisciplinary teams. The degree of configuration or interpretation required is also relevant. A tool capable of being used through a standard interface and ordinary professional training stands differently from one that requires extensive retraining, proprietary optimisation or intervention by a small group of specialists. Only capabilities that formed part of ordinary technical competence at the priority date should be attributed to the skilled person. The three considerations overlap and should not be applied mechanically. Their significance will vary according to the field, system, technical purpose and priority date. Relevant material may include licensing arrangements, open-source records, laboratory protocols, technical literature, professional guidance, training requirements, industry surveys and expert evidence of actual adoption. The party relying on the attribution of a particular AI tool should identify the factual basis for treating that tool or function as part of ordinary technical practice. Where validity is challenged, this ordinarily requires the party alleging obviousness to establish the relevant access, acceptance and deployability. An applicant or proprietor relying on an unexpected technical effect, unusual limitation or non-routine implementation should produce evidence supporting that assertion.207 Where expert evidence conflicts, the tribunal should ask whether each account reflects ordinary practice in the relevant field rather than the expert’s personal experience or the resources of an exceptional institution. Contemporaneous material should generally carry greater weight than retrospective assertion. Ordinary accessibility does not require universal access, but access confined to a small group of unusually well-resourced organisations should not be attributed to the skilled person unless those resources were normal for the practitioner or team represented by the benchmark.208

206

First Currency Choice (n 15) [28], [31]; Actavis (n 7) [59]–[60]. Towa Corp (n 127) [60]–[64]; IPOS, Examination Guidelines (n 15) paras 1.6–1.8, 2.24–2.32; EPO, Guidelines (n 15) G-IV, 7.5.3 and G-VII, 3.1. 208 First Currency Choice (n 15) [38]–[44], [50]; Mühlbauer (n 15) [47]–[48]; Actavis (n 7) [59]–[60], [88]. 207

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

Application of the Framework: AlphaFold

AlphaFold illustrates the need to distinguish scientific importance from incorporation into ordinary technical practice. Its significance to protein-structure prediction does not, without further evidence, justify attributing every version or function of the system to all biologists, medicinal chemists or drug-discovery teams.209 The analysis must remain specific to the field and the relevant date, with attention to whether practitioners could access the particular release, relied on it for the task in issue and possessed the expertise and resources needed to interpret its outputs. The operation of the framework may be illustrated by a hypothetical claim, with a priority date of 1 July 2022, to a molecule identified through AlphaFold-assisted structural analysis and computational screening. AlphaFold2 had by then been described in the scientific literature, and the AlphaFold Protein Structure Database had begun to make predicted protein structures available on a large scale. 210 Although those developments establish the availability of relevant technical resources, they do not, without evidence of practice in the particular field, demonstrate that AlphaFold had entered the skilled person’s ordinary technical environment for the target or research purpose in issue. The appropriate skilled team would first have to be identified. Depending on the nature of the claim, it might comprise a structural biologist, a medicinal chemist and a practitioner experienced in computational modelling. 211 The evidence would then need to establish whether the relevant AlphaFold release or database was realistically accessible to such a team, whether practitioners regarded it as sufficiently reliable for analysing the relevant class of targets, and whether effective use required only the data, computing resources and professional expertise ordinarily available in the field. Establishing those matters would place AlphaFold within the skilled team’s ordinary working methods, but it would not resolve the inventive-step inquiry. The analysis would still turn on whether the prior art would have led the team to investigate the target in that manner, the range

209

Jumper and others (n 1) 583–89; Varadi and others (n 2) D439–D444; General Tire (n 15) 482–83; EPO, Guidelines (n 15) G-VII, 3. 210 Jumper and others (n 1) 583–89; Varadi and others (n 2) D439–D444; AlphaFold Protein Structure Database, ‘AlphaFold DB’ (EMBL-EBI) <https://alphafold.ebi.ac.uk/> accessed 4 July 2026. 211 General Tire (n 15) 482–83; Mühlbauer (n 15) [20]–[21], [47]–[48]; EPO, Guidelines (n 15) G-VII, 3.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

and quality of the candidates generated, the technical judgment required to evaluate and select among them, and whether the claimed properties could reasonably have been anticipated before experimental confirmation. The weight attached to an AlphaFold prediction would depend on the circumstances in which it was produced and used. Relevant matters may include the confidence scores, known limitations of the predicted structure, the availability of homologous structures, the quality of the underlying sequence data, the number of plausible candidates generated and the extent of experimental work subsequently required.212 These factors would inform whether the prediction provided a sufficiently reliable basis for pursuing the claimed molecule and whether its asserted properties could reasonably have been expected. Even if AlphaFold formed part of ordinary practice, its availability would show only that the skilled team could have pursued that line of research. Obviousness would still depend on what the team would have been led to do, and reasonably to expect, in light of the prior art and the evidence. 213 The same principle applies to other AI tools. Their inclusion within the skilled person’s ordinary working methods does not make every resulting invention obvious. A claim may still depend on unusual problem framing, non-standard inputs, an unexpected technical property or a non-routine combination of computational prediction and experimental work. The availability of the tool is therefore only one part of the inventive-step inquiry; the claimed result must still be assessed on its own facts. D.

Common General Knowledge and Ordinary Technical Means

The argument rests on a distinction between what the skilled person knew and the means through which that knowledge could ordinarily be applied. Common general knowledge concerns the information and practices generally known and accepted as a sound basis for further work.214 The skilled-person inquiry extends beyond that body of knowledge, however, because the benchmark

212

Jumper and others (n 1) 583–89; Kathryn Tunyasuvunakool and others, ‘Highly Accurate Protein Structure Prediction for the Human Proteome’ (2021) 596 Nature 590, 590–96; Varadi and others (n 2) D439–D444; Actavis (n 7) [62]–[73]. 213 First Currency Choice (n 15) [44]–[51]; Actavis (n 7) [62]–[73], [88]; EPO, Guidelines (n 15) G-VII, 5.3. 214 General Tire (n 15) 482–83; Halliburton (n 62) [13].

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

is also attributed with the expertise, resources and capacity for routine experimentation normally available in the relevant field. 215 Applied to AI, the distinction separates knowledge of a tool from the practical ability to use it. Awareness of the system, its established functions, its recognised limitations and the purposes for which practitioners regarded it as reliable may form part of common general knowledge. Whether the skilled person could obtain access to the system and deploy it without exceptional datasets, infrastructure or specialist assistance is a related question concerning ordinary technical means. The evidence may overlap, but the two inquiries perform different functions. This approach avoids the artificial conclusion that software must itself be classified as knowledge before it can affect the obviousness analysis. It also reflects the broader doctrinal position across the three systems considered in this article. Singapore and UK law assess common general knowledge alongside the characteristics and practical capacities of the skilled person, while the EPO Guidelines expressly refer both to common general knowledge and to the means and capacity for routine work and experimentation normal in the field. 216 There is therefore no need to force every aspect of AI use into the doctrine of common general knowledge. The more accurate inquiry is whether the skilled person’s knowledge and ordinary working resources, taken together, reflected the technical practice of the field at the priority date. That inquiry can be accommodated within section 15 of the Singapore Patents Act, section 3 of the Patents Act 1977 and article 56 of the European Patent Convention. 217 Those provisions state the inventive-step standard without exhaustively defining the tools, methods or forms of expertise attributed to the skilled person, matters which have instead been developed through doctrine and evidence.218 The proposal accordingly requires no alteration to the statutory test. The skilled person remains a notional human practitioner, and the question remains whether the claimed invention would have

215

First Currency Choice (n 15) [28], [31]; Actavis (n 7) [59]–[60]; EPO, Guidelines (n 15) G-VII, 3. General Tire (n 15) 482–83; First Currency Choice (n 15) [28], [31], [38]–[44], [50]; EPO, Guidelines (n 15) GVII, 3. 217 Patents Act 1994 (2020 Rev Ed), s 15; Patents Act 1977, s 3; European Patent Convention, art 56. 218 First Currency Choice (n 15) [28], [31], [35]–[44]; Mühlbauer (n 15) [20]–[21], [47]–[48]; EPO, Guidelines (n 15) G-VII, 3. 216

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

been obvious at the priority date. The refinement lies in the factual construction of that practitioner: the attributed knowledge, resources and competence should correspond to the ordinary technical conditions under which work in the relevant field was then undertaken. E.

Hard Cases and Temporal Instability

The rapid development of AI creates an obvious difficulty for any framework tied to ordinary technical practice. Models may change within short periods, their capabilities may improve continuously, and tools regarded as important at one priority date may soon be superseded or abandoned.219 The problem is especially acute where a system becomes widely known before practitioners have acquired meaningful access to it, accepted it as sufficiently reliable or developed the competence required for its ordinary use. Patent law already fixes the inventive-step inquiry at the priority date. The rapid development of AI does not alter that principle, but makes contemporaneous evidence of actual practice especially important. Scientific prominence alone is insufficient. A system may be widely discussed yet remain inaccessible to ordinary practitioners, or used for preliminary screening without being relied upon for final technical decisions. A tool that had become routine by 2026 may not have occupied the same position in 2022.220 Evidence created after the priority date may assist in reconstructing earlier practice only where it reliably records conditions that already existed, such as earlier deployment, access arrangements or established professional use. Subsequent popularity, improved functionality or wider accessibility cannot retrospectively establish that the tool formed part of ordinary technical practice at the relevant date. Laboratory protocols, licensing records, technical guidance and

219

Abbott, ‘Everything Is Obvious’ (n 9) 35–39; Schellekens (n 9) 93–97; United States Patent and Trademark Office, ‘Request for Comments Regarding the Impact of the Proliferation of Artificial Intelligence on Prior Art, the Knowledge of a Person Having Ordinary Skill in the Art, and Determinations of Patentability Made in View of the Foregoing’ (2024) 89 Fed Reg 34217, 34217–18. 220 Patents Act 1994 (2020 Rev Ed), s 15; First Currency Choice (n 15) [41]–[44], [50]; EPO, Guidelines (n 15) GVII, 3, 5.1.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

evidence of actual adoption may therefore be used to test whether later assertions accurately describe the earlier position. 221 F.

Normative Objections

The proposed framework must nevertheless account for the continuing importance of human contribution, the disclosure function of the patent system and the commercial risks associated with AI-assisted research. Those considerations may explain why an AI-assisted invention is valuable, costly to develop or deserving of protection as a matter of policy. They do not, without more, establish that the claimed invention was non-obvious. Their relevance ultimately depends on whether they reveal a technical contribution extending beyond the knowledge and capabilities ordinarily attributed to the skilled person at the priority date. Human curation and validation present the most immediate difficulty. An output generated by an AI system will rarely pass directly into a patentable invention. Researchers may first have to define the technical problem, select and structure the relevant data, configure the model, interpret uncertain or incomplete outputs, eliminate false positives and identify candidates worthy of further investigation. Experimental work may then be required to confirm whether a predicted result possesses the technical properties on which the claim depends. 222 In many cases, these activities will involve substantial professional skill and may form an important part of the inventive process. Their presence does not, however, determine inventive step. The fact that a researcher expended considerable time, effort or resources after an AI system produced an initial result does not itself show that the claimed contribution went beyond ordinary technical practice.223 The significance of curation and validation depends on the nature of the judgment exercised. Inventive step may be supported where the researcher departed from an established workflow, recognised a technical significance that would not have been apparent to the ordinary skilled team, overcame a known limitation in the model or established an unexpected property through non-routine

221

First Currency Choice (n 15) [44]–[45]; Mühlbauer (n 15) [47]–[48]; Gregory N Mandel, ‘Patently NonObvious: Empirical Demonstration that the Hindsight Bias Renders Patent Decisions Irrational’ (2006) 67 Ohio State Law Journal 1391, 1421–23. 222 Shemtov (n 3) 7–9; Elton and others (n 3) 828–49; Butler and others (n 3) 547–55; Freilich and Rai (n 11) 924– 26. 223 First Currency Choice (n 15) [44]–[51]; Actavis (n 7) [62]–[73], [88].

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

experimentation. The position is different where selection, interpretation and testing were standard incidents of using the tool for the relevant purpose. The distinction is therefore not between inventions involving human intervention and those produced without it. Human involvement may be extensive in both routine and inventive work. What matters is whether that involvement supplied a technical contribution beyond the ordinary use of the AI system. A claim should not be regarded as inventive merely because a researcher reviewed an AI-generated output, selected one result from several candidates or subjected it to experimental confirmation. Equally, the use of AI should not obscure a genuinely inventive human contribution where the framing of the problem, interpretation of the output or design of the validation process went beyond what the ordinarily skilled team would have done. A separate objection arises from the disclosure function of patent law. AI-assisted research may produce useful technical information that an applicant would have little incentive to disclose without the prospect of exclusivity. 224 That consideration forms part of the traditional justification for the patent system, particularly where the information could otherwise be retained as confidential know-how. It cannot, however, be separated from the statutory conditions governing the grant of a patent. Patent protection is not conferred in return for disclosure alone. The disclosed subject matter must also be new, inventive and sufficiently disclosed. Where an AI-assisted output remains the product of ordinary technical practice, its usefulness or commercial sensitivity cannot supply an inventive step that is otherwise absent. To hold otherwise would risk conferring exclusionary rights over advances that practitioners, using the ordinary tools and methods of the field, might have produced without inventive ingenuity.225 That risk is especially significant in cumulative fields, where early patents may affect a wider programme of downstream research. Protection for routine AI-assisted outputs could restrict access to candidates, predictions or technical relationships that would otherwise have remained available as starting points for further investigation. The disclosure function remains important, but its justificatory force depends on the existence of a contribution satisfying the ordinary 224 225

Biogen (n 7) 34; Scotchmer (n 8) 30. Patents Act 1994 (2020 Rev Ed), ss 13–16, 25(4); Biogen (n 7) 34; Heller and Eisenberg (n 10) 698–701.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

requirements of patentability. It does not provide an independent reason to dilute the obviousness inquiry. Commercial risk and investment raise a further concern. In pharmaceuticals and other researchintensive sectors, the identification of a promising candidate is often followed by extensive laboratory validation, pre-clinical studies, clinical trials, regulatory review and manufacturing development. Those stages remain costly, lengthy and uncertain notwithstanding the use of AI at an earlier stage of discovery. 226 The prospect of patent protection may consequently be important to the commercial decision to pursue development. The economic importance of protection must nevertheless remain distinct from the technical question addressed by inventive step. A candidate may have been obvious to identify even though its subsequent validation and commercialisation required substantial expenditure. Conversely, a technically non-obvious invention may be relatively inexpensive to develop. The scale of downstream investment therefore does not reliably indicate whether the claimed solution went beyond ordinary skill.227 Allowing commercial risk to substitute for inventive step would also alter the character of the inquiry. Patent validity would come to depend on the cost of bringing an invention to market rather than on the technical contribution embodied in the claim. Investment may explain why exclusivity is commercially desirable, but it does not establish that the invention was non-obvious at the priority date. Human contribution, useful disclosure and commercial investment are therefore relevant within defined limits. They may support patentability where they reveal non-routine technical judgment, an unexpected technical property or some other contribution beyond ordinary practice. They should not, however, transform the routine product of an AI-assisted workflow into an inventive one merely because substantial effort, disclosure or financial risk followed its identification.

226 227

Freilich and Rai (n 11) 924–26; Actavis (n 7) [62]–[73]. First Currency Choice (n 15) [44]–[51]; Actavis (n 7) [62]–[73], [88].

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

The Preferred Position

The preferred approach is to retain the skilled person as a notional human practitioner, or team of practitioners, while allowing the knowledge and technical means attributed to that benchmark to reflect ordinary practice at the priority date. The benchmark should not be confined to an artificially analogue working environment once particular AI-assisted methods have become routine. Nor should it be endowed with frontier capabilities that remained inaccessible to ordinary practitioners or depended on exceptional expertise, data or infrastructure. An AI tool should therefore affect the inventive-step baseline only where the evidence establishes three matters. First, the relevant practitioner community could realistically access the particular tool or function. Secondly, practitioners ordinarily used it and regarded it as a sufficiently reliable basis for the task in issue. Thirdly, an ordinary practitioner or team could deploy it using the expertise, data and infrastructure normally available in the field. Awareness of the tool and acceptance of its established uses bear principally on common general knowledge. Access to the necessary software, data and infrastructure, together with competence in ordinary deployment, concerns the wider technical environment attributed to the skilled person. Attribution of the tool does not determine obviousness. It identifies one of the research routes available to the skilled person at the priority date. The tribunal must still ask whether the prior art would have led the skilled person to pursue that route, what technical choices and judgments the process required, what alternatives were available and whether the claimed result could reasonably have been expected. Routine access to an AI tool may therefore coexist with a non-obvious invention, just as uncertainty in a model’s precise output does not necessarily render inventive a result reached through an otherwise routine programme of work. This approach preserves the existing statutory framework. The skilled person remains human, the AI system remains an instrument and the legal question remains whether the claimed invention would have been obvious at the priority date. Its practical significance lies in ensuring that the inventive-step benchmark reflects ordinary research practice without allowing exceptional machine capability to replace ordinary technical skill.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

VI.

IMPLICATIONS FOR SINGAPORE

A. Why the Question Matters for Singapore The preceding analysis has wider comparative significance, but the issue carries particular importance for Singapore. Under the Singapore IP Strategy 2030, Singapore has sought to strengthen its position as a global hub for intangible assets, intellectual property, innovation and enterprise.228 Its research and innovation policies likewise place sustained emphasis on artificial intelligence, data, computing capacity, health and biomedical sciences, advanced manufacturing and the digital economy. 229 These are among the fields in which AI-assisted methods are most likely to become embedded in ordinary research practice. The connection with inventive step is direct. A skilled-person benchmark that disregards AI tools routinely used in those sectors may permit patent protection for results produced through ordinary technical work. A benchmark endowed with capabilities that remained beyond the practical reach of ordinary researchers may have the opposite effect, denying protection to inventions that continued to require substantial judgment, experimentation and technical risk. Either error would weaken the capacity of inventive step to distinguish genuine technical contribution from routine research practice. The concern is not simply whether Singapore grants too many or too few patents. The credibility of an IP system also depends on the consistency and intelligibility of the standards it applies. Applicants, investors and research institutions require a reasonably predictable account of the knowledge, resources and competence attributed to the skilled person. Uncertainty is likely to arise where decision-makers acknowledge that AI has altered research practice but lack a principled method for determining when a particular tool should affect the obviousness baseline.

228

Ministry of Law, ‘Strategy to Boost Singapore’s Position as a Global Intangible Assets & IP Hub Unveiled’ (26 April 2021) <https://www.mlaw.gov.sg/news/press-releases/2021-04-26-strategy-to-boost-singapores-position-as-aglobal-intangible-assets-n-ip-hub-unveiled/> accessed 10 July 2026; Intellectual Property Office of Singapore, ‘Singapore IP Strategy (SIPS) 2030’ (last updated 27 June 2025) <https://www.ipos.gov.sg/global-ip-hub/singaporeip-strategy-2030/> accessed 10 July 2026. 229 National Research Foundation, Research, Innovation and Enterprise 2030 (2025) 3–14, 30–31 <https://file.go.gov.sg/rie2030factsheet.pdf> accessed 10 July 2026; National Research Foundation, ‘Singapore Unveils S$37 Billion RIE2030 Plan to Advance Research and Innovation, and Support Singapore’s Economic and National Priorities’ (5 December 2025) 1–2 <https://www.nrf.gov.sg/files/RIE2030_Press_Release__website_.pdf> accessed 10 July 2026.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

The framework developed in this article responds to that problem without introducing a separate standard for AI-assisted inventions. It distinguishes what the skilled person ordinarily knew and accepted from the technical means through which that knowledge could be applied. In an AIrelated case, the analysis would therefore consider practical accessibility, routine use and accepted reliability, and ordinary deployability at the relevant priority date. Applied with appropriate evidential discipline, that approach may improve both patent quality and legal predictability in sectors that Singapore has identified as economically and strategically significant. B. Singapore’s Doctrinal and Institutional Position Singapore is well placed to develop this approach within its existing patent framework. Section 15 of the Patents Act states the inventive-step requirement by reference to the person skilled in the art, but does not exhaustively prescribe the knowledge, methods, resources or forms of expertise attributed to that person. 230 Those matters have been developed through judicial doctrine and evidence, with Singapore courts drawing extensively on UK patent authorities when identifying the skilled person, determining common general knowledge and applying the obviousness inquiry.231 This comparative relationship provides a substantial doctrinal foundation. Authorities such as General Tire, Pozzoli and Actavis support an inquiry that is sensitive to ordinary technical practice, the composition and competence of the skilled team, the research methods available at the priority date and the evidence concerning what practitioners would have done. Singapore need not create an AI-specific legal actor or alter the statutory test in order to recognise that ordinary research methods may change over time. The necessary refinement lies in the factual construction of the existing human benchmark. IPOS examination practice provides a complementary institutional basis. The Examination Guidelines apply established inventive-step principles and describe common general knowledge by reference to what practitioners generally knew and accepted as a sound basis for further action.232 The EPO Guidelines, although operating within a different legal system, offer a useful

230

Patents Act 1994 (2020 Rev Ed), s 15. First Currency Choice (n 15) [35]–[44]; Mühlbauer (n 15) [18]–[21]; Ng-Loy (n 26) paras 30.2.8–30.2.15. 232 IPOS, Examination Guidelines (n 15) paras 2.24–2.32, 4.18–4.24. 231

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

comparative distinction by referring separately to common general knowledge and to the means and capacity for routine work and experimentation normal in the field. 233 Together, these materials provide a vocabulary through which Singapore can distinguish knowledge of an AI tool from access to the system, the infrastructure required to operate it and the competence needed to use it effectively. The 2024 IPOS Supplemental Guidance for the Examination of AI-Related Patent Applications confirms that AI-related claims remain subject to the existing requirements of patentability. 234 Its focus is principally on the application of established principles to claims involving AI, rather than on the separate question of when an AI tool used during research should be attributed to the skilled person. The Guidance therefore supplies an institutional starting point but does not resolve the issue considered here. Further development would remain consistent with the cautious approach already adopted by IPOS. It would not require AI tools to be treated as common general knowledge merely because they were publicly available or technologically important. Nor would it treat the applicant’s use of AI as determinative. The inquiry would remain directed to the relevant technical community and to the position at the priority date. The respective roles of IPOS and the courts should nevertheless remain distinct. Examination guidance may promote consistency, identify relevant evidence and shape expectations before disputes arise. It cannot conclusively determine the content of section 15 or bind a court faced with contested evidence. Judicial development would therefore remain necessary where questions concerning the skilled person, common general knowledge or ordinary technical means arise directly in litigation. C. Implementation through IPOS Practice Examination practice provides the most immediate means of giving practical effect to the proposed framework. Judicial development depends on an appropriate dispute reaching the courts, whereas examination guidance can address recurring questions across a wider range of technologies. IPOS

233 234

EPO, Guidelines (n 15) G-VII, 3. IPOS, Supplemental Guidance (n 27) paras 2, 16–17.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

could therefore supplement its existing guidance with field-specific examples showing how the distinction between common general knowledge and ordinary technical means applies to inventions produced through AI-assisted research. The examples might address the use of protein-structure models, synthesis-planning systems, materials-discovery platforms or engineering-optimisation tools at particular priority dates. Their purpose would not be to classify individual systems conclusively as included within, or excluded from, the skilled person’s resources. AI models, versions and applications develop too quickly for fixed lists to remain useful. Worked examples would instead demonstrate how practical accessibility, routine use and accepted reliability, and ordinary deployability should be evaluated in context.235 That assessment should be grounded in evidence of practice within the relevant technical community. Depending on the field and the function attributed to the tool, relevant material may include licensing and access arrangements, open-source records, technical publications, professional guidance, laboratory protocols, computing and data requirements, training materials and documented patterns of adoption. 236 Publicity or isolated experimentation should not be equated with ordinary reliance, while commercial licensing should not be treated as the only means by which practical access may arise. The evidence may also inform the composition of the skilled team. AI-intensive research often combines domain knowledge with computational expertise, and the notional team should reflect that combination where it formed part of ordinary practice at the priority date. A medicinalchemistry team might include a practitioner experienced in computational modelling, while an advanced-manufacturing team might ordinarily draw on expertise in data analysis or optimisation. Specialist machine-learning capability should not, however, be attributed to every skilled team without evidence that such expertise was normally available in the relevant field. The applicant’s actual route to the invention must remain analytically distinct. Obviousness is assessed objectively, and the fact that an applicant used a particular AI system does not establish 235

General Tire (n 15) 482–83; First Currency Choice (n 15) [38]–[44], [50]; IPOS, Examination Guidelines (n 15) paras 2.24–2.32; EPO, Guidelines (n 15) G-VII, 3. 236 First Currency Choice (n 15) [38]–[44], [50]; Mühlbauer (n 15) [47]–[48]; IPOS, Examination Guidelines (n 15) paras 1.6–1.8, 2.24–2.32; United States Patent and Trademark Office, ‘Request for Comments’ (n 219) 34217–18.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

that ordinary practitioners could access or deploy the same tool for the relevant task. 237 IPOS should neither infer a lack of inventive step merely from the use of AI nor impose a general obligation to disclose AI involvement solely for that purpose. Where the status of the tool is material, the focus should remain on the knowledge, resources and practices of the relevant technical community at the priority date. Information about the applicant’s use of AI may nevertheless be relevant to other patentability requirements. Details concerning model configuration, training data, validation methods or the relationship between the AI system and the claimed technical effect may be necessary to establish support, sufficiency or a credible technical contribution. 238 Those matters concern whether the application adequately explains and supports the claimed invention; they should not displace the separate inquiry into the ordinary knowledge and technical means attributed to the skilled person. Guidance framed in these terms would not introduce a separate test for AI-assisted inventions. It would clarify the evidence relevant to the existing inventive-step inquiry, promote greater consistency in examination and reduce the risk that AI is either disregarded altogether or treated as decisive without an adequate factual basis. D. Regional and International Significance The implications of AI-assisted invention extend beyond domestic examination practice. Research, patent prosecution and commercialisation frequently take place across several jurisdictions, while inventive step remains governed by national or regional patent law. Material differences in the treatment of artificial intelligence (“AI”) within the skilled-person inquiry may therefore produce inconsistent outcomes for related inventions, increase prosecution costs and reduce predictability for applicants operating internationally.239 Singapore occupies a useful position within this developing discussion. Its inventive-step doctrine draws substantially on common-law authorities concerning the person skilled in the art and 237

Windsurfing (n 5) 73–74; Pozzoli (n 5) [23]; First Currency Choice (n 15) [41]–[45]; Actavis (n 7) [59]–[60]. IPOS, Supplemental Guidance (n 27) paras 8–10, 16–17; T 0702/20 (n 29) [12]–[19]; T 1425/21 (n 164) [15]– [20.2]. 239 World Intellectual Property Organization, ‘Artificial Intelligence and Intellectual Property Policy’ (WIPO) <https://www.wipo.int/en/web/frontier-technologies/artificial-intelligence/conversation> accessed 10 July 2026; United States Patent and Trademark Office, ‘Request for Comments’ (n 219) 34217–18. 238

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

common general knowledge, while the Intellectual Property Office of Singapore (“IPOS”) operates within an internationally connected examination system and engages with developments in other major patent offices. Singapore also participates in regional patent work-sharing through the ASEAN Patent Examination Co-operation (“ASPEC”).240 These features allow Singapore to draw together doctrinal reasoning and administratively workable forms of technical assessment. They do not, however, give Singapore authority to harmonise inventive-step doctrine across jurisdictions. Any approach adopted domestically would remain subject to the statutory framework of the Singapore Patents Act, while other patent offices would continue to apply their own legislation and examination practices. Singapore’s contribution would instead lie in articulating a method that is sufficiently clear to be understood across systems: common general knowledge would identify what practitioners ordinarily knew and accepted, while practical accessibility, accepted field-level use and ordinary deployability would inform the wider technical means attributed to the skilled person. ASPEC provides one channel through which the quality of Singapore’s examination practice may acquire wider regional relevance. It does not establish a unitary Association of Southeast Asian Nations patent system or require participating offices to adopt one another’s conclusions.241 Search and examination results may nevertheless assist the prosecution of corresponding applications in other participating jurisdictions. Clear reasoning on the role of AI in inventive step may therefore inform regional practice indirectly, although the final determination would remain with each national patent office. The wider international discussion confirms that the issue is no longer speculative. The World Intellectual Property Organization (“WIPO”) has examined the implications of AI for intellectual property, while patent offices in other jurisdictions have considered how AI may affect prior art, the level of ordinary skill in the art and determinations of patentability. 242 Singapore’s contribution would lie not merely in recognising that AI affects the inventive process, but in developing a 240

Intellectual Property Office of Singapore, ‘ASEAN Patent Examination Co-operation’ (last updated 28 April 2026) <https://www.ipos.gov.sg/about-ip/patents/how-to-register-overview/acceleration-programmes/asean-patentexamination-cooperation/> accessed 10 July 2026. 241 Intellectual Property Office of Singapore, ASEAN Patent Examination Co-operation Programme: Notice and Procedures (rev August 2023) paras 1.1–1.4; IPOS, ‘ASEAN Patent Examination Co-operation’ (n 240). 242 World Intellectual Property Organization, ‘Artificial Intelligence and Intellectual Property Policy’ (n 239); United States Patent and Trademark Office, ‘Request for Comments’ (n 219) 34217–18.

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

principled method for determining when a particular AI tool had entered ordinary technical practice at the relevant priority date. That contribution should remain measured. The prominence of an AI system should not, by itself, justify its attribution to the skilled person, nor should an invention be presumed obvious merely because AI contributed to its development. The inquiry should remain grounded in evidence of the practice of the relevant technical community, including whether practitioners could obtain meaningful access to the system, relied upon it for the task in issue and could deploy it using ordinary expertise and resources. A framework expressed in those terms would serve both domestic and comparative purposes. Within Singapore, it could improve the consistency and technical credibility of examination by reducing the risk that AI is either disregarded altogether or treated as decisive without an adequate evidential basis. Beyond Singapore, the same reasoning may provide a useful point of comparison for patent offices confronting similar questions, although its persuasive value would depend on the clarity of the analysis rather than any formal regional effect. The principal benefit nevertheless lies in the quality and predictability of Singapore’s own patent system. Its position as an intellectual-property hub depends not only on efficient protection, but also on confidence that granted rights correspond to genuine technical contributions. A more precise account of the skilled person’s ordinary knowledge and technical means would support that objective while allowing Singapore to address AI-assisted invention through established inventive-step doctrine rather than a separate regime for AI. VI.

CONCLUSION

This article began with a growing mismatch between inventive-step doctrine and contemporary research practice. Inventive step continues to ask what would have been obvious to the person skilled in the art, yet ordinary work in structural biology, pharmaceuticals, materials science and advanced engineering is increasingly shaped by artificial intelligence. The familiar debate over whether an AI system may qualify as an inventor captures only part of the difficulty. The more immediate pressure arises where the inventor remains human, but the route to the invention is

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

materially affected by tools that practitioners in the relevant field may already use as part of ordinary research. The law should neither disregard those tools nor allow the capabilities of frontier AI to define the skilled person. A benchmark that excludes AI systems which had entered ordinary practice may understate the knowledge and technical capacity of the relevant practitioner community. The opposite approach would be equally distortive. Attributing unrestricted machine capability to the skilled person would replace ordinary technical competence with an idealised computational standard that few, if any, practitioners could realistically exercise. The first approach risks treating the products of routine AI-assisted work as inventive, while the second may deny protection where the claimed advance continued to depend on substantial technical judgment, experimentation and uncertainty. The more defensible position is to retain the skilled person as a notional human practitioner, or team of practitioners, while ensuring that the knowledge and technical means attributed to that benchmark correspond to ordinary practice at the priority date. Common general knowledge identifies what practitioners ordinarily knew and accepted as a sound basis for further work. A related inquiry concerns the resources, expertise and capacity for routine experimentation normally available to them. An AI tool may therefore affect the inventive-step baseline where practitioners knew of and accepted it for the relevant purpose, could obtain meaningful access to it and could deploy it using the expertise and resources ordinarily available in the field. Attribution of the tool does not create a presumption that an AI-assisted invention is obvious. It establishes part of the technical setting within which obviousness must be assessed. A claim may remain non-obvious because the prior art would not have led the skilled team to pursue the relevant route, because the claimed result could not reasonably have been expected, or because successful selection, interpretation or implementation depended on judgment extending beyond the ordinary use of the tool. Conversely, uncertainty as to the precise output does not necessarily render inventive a result reached through a routine and predictable programme of work. The typology developed in this article clarifies the scope of that argument. Autonomous AI invention presents the most dramatic conceptual challenge, while cases in which AI contributes materially to the inventive concept may continue to raise difficult questions of attribution and 95

Singapore Comparative Law Review

Rethinking The Person Skilled In The Art In The Age Of Artificial Intelligence

inventorship. The immediate inventive-step problem, however, arises most clearly where AI operates as a research tool. In those cases, the central issue is whether the tool had become part of the skilled person’s ordinary technical environment, rather than whether the tool itself should be recognised as an inventor. The proposed framework does not resolve every boundary case, particularly where computational and human contributions cannot readily be separated, but it provides a principled basis for addressing the cases already emerging in ordinary research practice. For Singapore, the proposal offers a measured response within the existing statutory framework. The Intellectual Property Office of Singapore may clarify the evidence relevant to practical accessibility, accepted field-level use and ordinary deployability, while the courts may continue to develop the skilled-person inquiry through established principles of common general knowledge, ordinary competence and the avoidance of hindsight. Such an approach would support greater consistency and patent quality in AI-intensive sectors without requiring a separate regime for AIassisted inventions. It may also provide a useful point of comparison for other jurisdictions confronting the same problem. Inventive step need not be reconstructed around machine intelligence. Existing doctrine is capable of responding to changes in technical practice, provided that the skilled person is constructed from an accurate account of the work ordinarily undertaken in the relevant field. AI should not be ignored once it has become a dependable and routinely used research tool, but neither should its full capabilities be substituted for the human judgment represented by the skilled person. Ordinary skill may evolve with ordinary tools without making every result produced through those tools obvious.

PRACTICE MADE PERFECT. Start an altogether exceptional career with us.

We are accepting applications for relevant legal training in 2027 and internship positions in 2026. For more information, please visit www.drewnapier.com/Careers.

Singapore Comparative Law Review

Chaos In Action

CHAOS IN ACTION: REASSESSING SINGAPORE’S APPROACH TO CRYPTOASSET CLASSIFICATION AND DEFINING THE BOUNDARIES OF THE TERTIUM QUID Davon Pung*, Shanice Yee** Honourable Mention for Best Article Award 2026 ABSTRACT Personal property rights have hitherto been defined by the traditional binary between choses in action and possession inherited from the judgment in Colonial Bank v Whinney (1885) 30 Ch D 261 (CA). In Singapore, courts have recognised or assumed that cryptoassets are capable of attracting proprietary rights without definitively classifying them. Going forward, it becomes increasingly pertinent to establish how Singapore may identify the assets warranting third-category treatment. In Bybit Fintech Ltd v Ho Kai Xin and others [2023] SGHC 199 (“Bybit”), the High Court classified the stablecoin USDT as a chose in action. Bybit’s reasoning draws on Holdsworth’s account of the diverse incorporeal property historically classed as choses in action, arguing that the category remains flexible and open. However, this wide conception reduces choses in action to a catch-all, obscuring the legal incidents organised by the undistended orthodox classification. This article contends that neither choses in action nor choses in possession comfortably accommodates cryptoassets that function independently of obligations owed by an identifiable counterparty: the incidents of choses in possession are calibrated to tangible things and the narrow conception of a chose in action presupposes an enforceable right. In England, Wales and Northern Ireland, the Property (Digital Assets etc) Act 2025 (“P(DA)A 2025”) endorses a view that falling outside that binary does not itself preclude an asset from personal property rights. This produces a deliberate minimalism that contrasts with Singapore’s incidental drift in cryptoasset jurisprudence. This article advances that Singapore should recognise third-category assets statutorily while providing greater guidance than the P(DA)A 2025 model. This article proposes a two-layer classificatory framework. The first layer identifies paradigm third-category assets through four core indicia: rivalrousness, independence from the legal system, control

University College London (LLB). Class of 2027. University College London (LLB), Columbia University (JD). Class of 2028. We are grateful to the editorial team for their thoughtful and constructive comments on earlier drafts of this article; any remaining errors are ours. **

Singapore Comparative Law Review

Chaos In Action

through cryptographic mechanisms and the absence of a counterparty obligor. The second layer provides two supplementary indicia—susceptibility to unauthorised factual interference and persistence of identity through transfer—for borderline cases. The proposed changes supply a principled basis for decision-making which supports certainty and preserves much-needed adaptability in this field characterised by rapid destabilising developments. I.

INTRODUCTION

Singapore’s cryptoasset jurisprudence has developed incrementally, through a varied sequence of appellate, interlocutory and summary decisions in which the courts have been willing to protect particular cryptoassets as objects of proprietary claims without producing a settled personal property taxonomy. Courts in Singapore have recognised or assumed that some cryptoassets are capable of attracting proprietary protection but have not explained which legal consequences ought to follow from an asset’s classification or how that classification should consistently be reached. The clearest classificatory statement came in Bybit Fintech Ltd v Ho Kai Xin and others,1 where the High Court, on an application for summary judgment concerning the stablecoin USDT, treated the asset as a chose in action because the category’s historical expansion and the resulting diversity of the things it contains show it to be “broad, flexible, and not closed”. 2 For all its reliance on Holdsworth’s history of the category, the reasoning departs from Holdsworth’s own conclusion that its unwieldy extension had rendered the modern law of choses in action unsatisfactory. 3 This article argues that cryptoassets whose operation depend on no enforceable claim against an identifiable counterparty fall outside of choses in action in its narrow sense, and that the wide conception capable of accommodating them should be rejected. The wide conception dilutes the defining features of choses in action while requiring a sub-classification indistinguishable in substance from the third category it is invoked to avoid. This article proposes that Singapore go further than the United Kingdom’s Property (Digital Assets etc) Act 2025 (“P(DA)A 2025”) by enacting as legislation or statutory guidance two layers of indicia that guide courts in identifying assets qualifying for third-category treatment. Their incidents will be left to be refined by the

[2023] SGHC 199. Ibid, [35]. 3 WS Holdsworth, ‘The History of the Treatment of Choses in Action by the Common Law’ (1920) 33 Harvard Law Review 997, 1029. 2

Singapore Comparative Law Review

Chaos In Action

common law, allowing legislative recognition to add clarity without imposing the rigidity of a static regime. Much of the existing scholarship has focused on two anterior questions of whether cryptoassets can be property and whether the common law should recognise a third category of personal property. This article addresses a narrower, practical subset of those questions: how Singapore should identify the assets that qualify for third-category treatment. Three principal contributions are advanced to answer this question. First, it is argued that neither limb of the orthodox binary accommodates qualifying cryptoassets without distortion. Second, it is posited that from that conclusion follows a case for statutory recognition in Singapore, informed by the statutory model now in force in England, Wales and Northern Ireland and by its early treatment in English courts. Third, the article proposes a two-layer framework in which four core indicia identify the paradigm third-category asset and two supplementary indicia guide judicial determination where an asset lies at the margins. The constituent indicia, drawn from the Law Commission’s analysis and precedents in common law, serve to structure the next generation of cryptoasset litigation in a manner adapted to Singapore’s institutional setting. The analysis that follows is framed by a necessary premise that cryptoassets are not excluded from proprietary recognition as a class. Depending on their architecture, particular cryptoassets may be objects of personal property rights, a proposition that English and Singapore courts have increasingly accepted in relation to cryptocurrencies and non-fungible tokens by reference to the long-standing National Provincial Bank v Ainsworth criteria.4 Further, as Chan observes, proprietary recognition is only the beginning. It remains necessary to explain how and why “particular rules of title originally developed in the context of tangible property should apply” to blockchain technology.5 Throughout, a distinction is maintained between the asset and the legal rights arising in relation to it is maintained throughout. The article proceeds in four main sections. The remainder of Section I defines the assets under consideration and establishes the proprietary baseline from which the classificatory analysis may begin. Section II examines the orthodox binary of choses in action and in possession, encapsulated

4 5

[1965] AC 1175 (HL). Timothy Chan, 'The Nature of Property in Cryptoassets' (2023) 43 Legal Studies 480, 480.

Singapore Comparative Law Review

Chaos In Action

in Colonial Bank v Whinney, arguing that it was fashioned for another age and cryptoassets do not belong in either category. 6 Section III sets out the comparison on which our proposal rests, which contrasts Singapore’s incidental drift in common law with the deliberate statutory minimalism adopted in England, Wales and Northern Ireland. Section III also examines the consequences of leaving classification unresolved, turning to the early post-Act decision of Ping Fai Yuen v Fun Yung Li & Another and the pertinent discourse on the persistence of cryptoasset identity through transfer.7 Section IV sets out the two-layer framework and addresses its principal objections. Without endorsing the proliferation of cryptoassets, this article recognises the inevitable litigation that may sprout from this nascent technology. Classification supplies an organising premise for the development of those rules. It does not itself govern proprietary remedies, the treatment of client assets on custodian or exchange insolvency, transfer of title, or liability for wrongful interference; rather, it identifies the analogies from which those incidents should be developed. The proposed taxonomic framework would provide courts with a defensible foundation to develop those rules and reach more principled and predictable conclusions than the present dichotomy allows for, while retaining sufficient flexibility to accommodate the continuing evolution of blockchain and distributed ledger technology. A. Key definitions The “comparative lack of attention to classification, and relative paucity in vocabulary for discrete concepts” within common law systems has caused considerable confusion in cryptoasset scholarship.8 Precise working definitions are therefore essential. Beginning with the broadest relevant term, digital assets, the UK Law Commission adopts a pragmatic catch-all, covering any asset represented digitally or electronically while recognising that not all attract personal property rights.9 Singapore’s Payment Services Act 2019 likewise avoids a static definition, by opting for an activity- rather than asset-based classification, apt for regulatory purposes but carrying no implication for proprietary status. Section 2(1) of the Payment Services Act 2019 defines a “digital

(1885) 30 Ch D 261 (CA). [2026] EWHC 532 (KB). 8 Kelvin FK Low and Megumi Hara, ‘Cryptoassets and Property’ in Sjef van Erp and Katja Zimmermann (eds), Research Handbook on EU Property Law (Edward Elgar 2024) 146, 146. 9 Law Commission, Digital Assets: Final Report (Law Com No 412, 2023), 9. 7

100

Singapore Comparative Law Review

Chaos In Action

payment token” by reference to its function as a medium of exchange, 10 thus excluding many cryptoassets such as utility tokens and NFTs.11 The article accordingly adopts the Law Commission’s broad formulation as a baseline before narrowing the discussion to cryptoassets, which raise the most acute difficulties under orthodox personal property law. We use ‘cryptoasset’ in a broad sense consistent with case law and commercial usage, to refer to blockchain-instantiated tokens including Bitcoin, Ether and NFTs. The Law Commission’s Digital Assets: Final Report uses the term more narrowly, reserving ‘cryptoasset’ for a crypto-token stapled to a legal right or interest in another thing, such as tokenised securities, central bank digital currencies and redeemable stablecoins.12 ‘Crypto-token’ is then used for things denominated in their own notional unit of account, otherwise referred to as unbacked. The Commission notes that algorithmic stablecoins, being stapled to no legal right, are better described as crypto-tokens.13 Here, ‘cryptoasset’ is used in the broader sense and the backed/unbacked terminology is adopted where the distinction is pertinent. Whether the token-holder’s position rests on an enforceable right against an identifiable obligor or solely on the token’s architecture is an important distinction to which we return when addressing indicium (d) in Section IV. Tokens embodying the former rights, of which tokenised debt securities are the clearest case, are more readily accommodated as choses in action. Cryptoassets carrying no stapled legal right, whether fungible or non-fungible, form the primary focus of the following discussion. B. Establishing a baseline of proprietary rights for cryptoassets Before determining which category of personal property most readily accommodates cryptoassets, it is necessary to establish that they are capable of constituting objects of personal property as a generalised class. The proprietary baseline is supplied by the reasoning in AA v Persons Unknown, where Bryan J, relying on the UK Jurisdiction Taskforce’s Legal Statement, held that a cryptotoken could be an object of personal property rights even if it was not a chose in action in the narrow sense, describing as fallacious the contrary assumption that the law recognises no form of

Payment Services Act 2019, s 2(1). Financial Conduct Authority, Guidance on Cryptoassets: Feedback and Final Guidance to CP 19/3 (PS19/22, July 2019) para 2.21; Financial Services Consumer Panel, Annual Report 2023 (6 September 2023) 23. 12 Law Commission (n 9), para 8.12(2) and Glossary. 13 Ibid, fn 856. 11

101

Singapore Comparative Law Review

Chaos In Action

property beyond choses in possession and choses in action. 14 The Legal Statement utilised Lord Wilberforce’s indicia in Ainsworth as the foundation for determining if a thing is property: that they are definable, identifiable by third parties, capable in their nature of assumption by third parties, and have some degree of permanence. 15 Proprietary protection has become increasingly important as cryptoasset use grows in both the UK and Singapore, with the costs of indeterminacy borne by both sophisticated market participants and ordinary consumers. The UK Jurisdiction Taskforce supported the Ainsworth approach “in circumstances where there is ample evidence that there is a large and active market in which cryptoassets are acquired and traded as things of value”, applicable to both aforementioned jurisdictions.16 Singapore has among the world’s highest cryptoasset ownership rates, reported at 26% of residents in an industry survey. 17 Singapore’s subordinate legislation reflects this approach. Order 22 r 1(1) of the Rules of Court 2021 defines movable property to include “cryptocurrency or other digital currency”, which Jeyaretnam J in ByBit described as express recognition of cryptocurrency as a form of property capable of being the subject matter of an enforcement order. 18 The significance of the classification is most evident on the insolvency of a cryptoasset exchange. If customers retain a proprietary interest in cryptoassets held by the exchange as custodian or trustee, they may generally claim those assets outside the intermediary’s insolvent estate, subject to questions of segregation, pooling and identification. By contrast, if holders only have contractual rights, they rank pari passu with other unsecured creditors and may recover only a fraction of their holdings’ value.19 As the UK Jurisdiction Taskforce has observed, the general

[2019] EWHC 3556 (Comm), [2020] 4 WLR 35 [55]–[61]. UK Jurisdiction Taskforce, Legal Statement on Cryptoassets and Smart Contracts (LawTech Delivery Panel 2019) paras 39–40; Ainsworth (n 4) 1248. 16 UK Jurisdiction Taskforce (n 15), 40. 17 Sheila Chiang, ‘More People Own Crypto in Singapore, with Younger Users Leading the Way: Survey’ The Straits Times (8 April 2025) <https://www.straitstimes.com/business/more-people-own-crypto-in-singapore-withyounger-users-leading-the-way-survey> accessed 3 July 2026. 18 Bybit (n 1) [30] 19 Paúl Noboa-Velasco, “‘Crypto-Custodians’ Insolvency Proceedings: Clarifying the Proprietary Nature of cryptoassets’” (Singapore Global Restructuring Initiative Blog, 5 September 2023) <https://ccla.smu.edu.sg/sgri/blog/2023/09/05/crypto-custodians-insolvency-proceedings-clarifying-proprietarynature-crypto> accessed 3 July 2026. 15

102

Singapore Comparative Law Review

Chaos In Action

rules of tracing can be adapted to digital assets without fundamental modification, though application will depend on the nature of the specific assets in question. 20 The article now turns to the critics of proprietary recognition, of whom Robert Stevens is among the most prominent. He raises three principal objections: (i) that cryptoassets cannot be physically possessed; (ii) that holders lack a “cause of action to enforce a primary right” because no tort, contractual or statutory entitlement exists; and (iii) cryptoassets are “unproductive” and “positively harmful”, facilitating environmental pollution and criminality, such that their use should be eliminated where possible. 21 We do not find these arguments convincing. Objection (i) assumes that property must be capable of physical possession. Even if cryptoassets cannot be physically possessed, they may still constitute choses in action or, as we argue, members of a third category. Another possibility, which we ultimately reject in Section 2.2, is to classify them as choses in possession, a category arguably no longer defined by physical control. 22 Objection (ii) is more substantial as Stevens correctly identifies why a native decentralised token falls outside the narrow sense of a chose in action, a proposition we accept and develop in Section 2.3. It does not follow, however, that such an asset is incapable of being property. As recognised in AA, proprietary capacity does not depend on an asset being a chose in action in that sense. 23 The objection assumes the exhaustiveness of the orthodox binary, the very issue in contention, and is difficult to reconcile with the trajectory of Singapore law. Although Singapore’s courts have not finally resolved the classificatory question, they have recognised proprietary claims in relation to cryptoassets. Janesh s/o Rajkumar v Unknown Person (“CHEFPIERRE”) recognised a serious question to be tried as to whether an NFT was property capable of supporting proprietary relief, 24 while ByBit held that the holder of a cryptoasset has “in principle an incorporeal right of property

UK Jurisdiction Taskforce, Legal Statement on Digital Assets and English Insolvency Law (LawtechUK, April 2024) 35. 21 Robert Stevens, ‘Crypto Is Not Property’ (2023) 139 Law Quarterly Review 615, 621. 22 Lucas Clover Alcolea and Jan Mihal, ‘The Tiptoe to Crypto: An Analysis and Account of Property in Cryptocurrency’ (2025) 54 Common Law World Review 46, 61. 23 AA (n 14) [55], [58]–[59]. 24 [2022] SGHC 264.

103

Singapore Comparative Law Review

Chaos In Action

recognisable by the common law as a thing in action and so enforceable in court”, granting a declaration of constructive trust on that footing. 25 Objection (iii) concerns the desirability of cryptoassets, a question analytically distinct from whether they are capable of being property. Property law identifies the objects capable of ownership; Parliament may then regulate, tax or prohibit where public policy requires. As Lord Millett opined in Foskett v McKeown, property rights “are determined by fixed rules and settled principles” and concepts of fairness “which in reality mask decisions of legal policy, have no place in the law of property”.26 Numerous forms of property such as tobacco, hazardous chemicals, and pollutants are harmful, but that has never been treated as a reason to deny their proprietary character. II.

FAILURE OF THE ORTHODOX BINARY

A. Colonial Bank v Whinney and the orthodox binary Personal property has historically been divided into two categories of choses in possession and choses in action and treated as exhaustive. Fry LJ laid the groundwork in Colonial Bank: “all personal things are either in possession or in action. The law knows no tertium quid between the two”.27 The UK Law Commission summarises a thing in possession as any object “amenable to possession”, including assets which are “tangible, moveable and visible”. 28 Meanwhile, the second category comprises choses in action, commonly described in a narrow sense as “rights in things in action, denied physical enjoyment, are asserted by taking legal action or proceedings”. 29 Common examples include debts, rights to sue for breach of contract, and company shares. This classification is not merely semantic. As recognised by Akber Datoo, an asset’s classification informs the legal rights its owner enjoys and “uncertain characterisation moves itself into

ByBit (n 1) [36]. [2000] UKHL 29, [2001] 1 AC 102, 127 (Lord Millett). 27 Colonial Bank v Whinney (n 6) at 285. 28 Law Commission (n 9) para 3.17. 29 Michael Bridge, Louise Gullifer, Kelvin FK Low and Gerard McMeel, The Law of Personal Property (3rd edn, Sweet & Maxwell 2021) para 4.002. 26

104

Singapore Comparative Law Review

Chaos In Action

uncertainty thereafter”. 30 The orthodox binary, a creature of historical distinctions, has struggled to accommodate cryptoassets. This article argues that cryptoassets are neither choses in possession nor choses in action, but belong within a third category of personal property. B. Cryptoassets are not choses in possession This section rebuts arguments that cryptoassets should fall within the category of choses in possession on the ground that it unduly distorts the category’s established limits. Alcolea and Mihal advance two arguments: first, that possession should not be limited to physical control; second, that other jurisdictions have already extended it to certain cryptoassets. 31 Tangibility First, they submit that technological developments permit comparable practical control over intangible and tangible assets, rendering the requirement of physical control obsolete. Honoré’s broader account of possession as “exclusive physical control of a thing, or … such control as the nature of the thing admits” lends the argument intuitive force, 32 an observation buttressed by technological developments that challenge the traditional notions of tangibility. 33 Traditionally, possession comprises both a mental and physical element. 34 The court’s formulation in R v Boyesen that what is required is “physical control or custody of a thing”35 resonates with the approach later taken in JA Pye (Oxford) Ltd v Graham36 and OBG v Allan37 that this will provide evidentiary certainty.38 An intangible asset thus cannot be subject to possession. Possession is also far less certain for cryptoassets than tangible assets. For a cryptoasset, the position can be more complicated: situations where two people possess copies of the same private key, multisignature wallets where two people are required to authorise a transaction and smart contracts are all 30

House of Lords, Property (Digital Assets etc) Bill [HL] Special Public Bill Committee: Corrected Oral Evidence: Property (Digital Assets etc) Bill [HL] (5 December 2024) Q21. 31 Clover Alcolea and Mihal (n 22) 55. 32 AM Honoré, ‘Ownership’ in AG Guest (ed), Oxford Essays in Jurisprudence (Clarendon Press 1961) 107, 113. 33 Jessica Lai, ‘Creating Coherency in Conversion of (In)Tangible Property in New Zealand’ (2021) 28 Tort Law Review 36. 34 John W Salmond, Jurisprudence (Glanville L Williams ed, 12th edn, Sweet & Maxwell 1966) 321. 35 [1982] AC 768, 768. 36 [2002] UKHL 30, [2003] 1 AC 419. 37 [2007] UKHL 21, [2008] 1 AC 1. 38 Graham (n 36).

105

Singapore Comparative Law Review

Chaos In Action

examples that undermine any straightforward analogue to possession. While possession could be redefined to encompass these arrangements, doing so would depart unjustifiably from its traditional meaning historically centred on its physical nature. The Law Commission’s preference for a bespoke concept of control for third category assets rather than extending possession is accordingly preferable.39 Other Jurisdictions Alcolea and Mihal argue that possession has expanded to encompass custody and control in other jurisdictions, outside property law. Examples include the New Zealand tort law case of Henderson v Walker;40 the Court of Appeal of England and Wales case of R v Cyprian Okoro (No 3) in interpreting the Criminal Justice Act 1988 as applied to digital images; 41 and R v Morelli,42 a Canadian criminal law case. They therefore argue that cryptoasset holders exercise custody and control because private keys enable transfer while excluding others’ access to the asset. They contend that this is the “digital equivalent of [the first limb of] ‘possession’”.43 This analogy nevertheless overlooks that the incidents of choses in possession remain calibrated to tangible assets. This is reflected in remedies such as conversion, bailment and pledges. Turning to the tort of conversion, extending it to cryptoassets would introduce manifest legal uncertainty. In Singapore, Alwie Handoyo v Tjong Very Sumito confirmed that conversion only applies to tangible property, subject to the limited exception of documentary intangibles, which are in essence tangible representations of choses in action.44 Unlike tangible goods which are interfered with through acts such as taking, detention or destruction, cryptoassets do not have tangible representations. They are interfered with through unauthorised transfers within a distributed ledger or compromised cryptographic credentials and their functionality already eschews possession. Extending conversion to accommodate these fundamentally different forms of interference would require the tort to evolve beyond its traditional possessory rationale rather than merely applying established principles to a new factual context.

Law Commission (n 9) chapter 5. [2019] NZHC 2184 [248–249]. 41 [2018] EWCA Crim 1929, [2019] 1 WLR 1638, 1647. 42 [2010] 1 SCR 253. 43 Clover Alcolea and Mihal (n 22), 62. 44 [2013] SGCA 44, [2013] 4 SLR 308 [131]. 40

106

Singapore Comparative Law Review

Chaos In Action

Conversion’s nature as a strict liability tort raises further difficulties. In particular, the Law Commission warned that “the different properties, behaviours and characteristics of third category things might make it harder for people innocently engaging with distributed systems to avoid accidentally interfering with a claimant’s object of personal property rights”. 45 Increased unexpected tortious liability may precipitate a chilling effect. The City of London Law Society likewise cautioned that this could create unacceptable legal risk for “operators/administrators of private, permissioned blockchain or DLT-based systems for the holding and transfer of such digital assets” and “reduce the UK’s attractiveness as a governing jurisdiction for such systems”. 46 It might be argued that cryptoassets could be treated as choses in possession subject to discrete asset-specific exceptions (e.g. a carve-out where they are not capable of conversion). However, the more carve-outs cryptoassets require, the less useful the classification becomes. Conversely, applying physical possession-based mechanisms such as conversion, bailments, liens and pledges would rewrite a substantial body of private law. In either case, cryptoassets effectively become a new category concealed within the old category of choses in possession. Where an entirely new body of principles must be created to accommodate an asset, recognising a distinct category is conceptually cleaner than disguising that reconstruction within the language of possession. A third category permits the direct and selective refinement of these incidents and principles without distorting the established meaning of possession. C. The doctrinal costs of expanding choses in action Some commentators advocate a broad conception of choses in action as a residual category encompassing all personal property “that is not a thing in possession”.47 This article contends that this reduces choses in action to a catch-all, undermining the distinctions the classification is intended to draw. Absent a third category, courts in Singapore have treated cryptoassets as choses in action even where they do not satisfactorily meet the criteria. The Narrow Conception

Law Commission (n 9) para 9.73. Ibid at footnote 1200, quoting City of London Law Society Financial Law Committee, response to Digital Assets: Consultation Paper (2022). 47 Law Commission (n 9) para 3.18. 46

107

Singapore Comparative Law Review

Chaos In Action

In its earliest sense, a chose in action referred to “things in respect of which a man had no actual possession or enjoyment, but a mere right enforceable by action”. 48 Cryptoassets do not satisfy this definition because, as most Law Commission consultees observed, they are not ‘rights or claims in themselves’ and exist independently of enforceable legal claims. 49 This may differ in private, permissioned systems, or intermediated holdings, involving a central actor (e.g. a platform operator, custodian or other intermediary) against whom cryptoassets in the form of enforceable rights may exist. However, these exceptional cases should not be generalised to the great majority of decentralised cryptoassets, such as Bitcoin and Ether. Decentralised assets’ value and operation do not depend upon any corresponding obligation owed by an identifiable counterparty. Further, Holdsworth explains that because medieval law was remedies-based, the distinction is not just conceptual but procedural. Actions to recover physical things (such as through replevin, detinue and trespass) gave rise to choses in possession and actions to enforce rights gave rise to choses in action.50 This supports the view that any enforceable rights arise from the holding arrangement rather than the cryptoasset itself. The Wide Conception and the Pitfalls of a Sub-Classificatory System Holdsworth attributes the wide conception to the “peculiar division of English law into common law and equity”. 51 Choses in action, over time, have expanded in scope, coming to “include this great mass of miscellaneous rights”. 52 Under this view, choses in action now constitute a residual category encompassing everything that is not a chose in possession. Holdsworth argues that the diversity of judicial treatment for current choses in action necessarily produces diverse legal incidents, such that in practice, they are governed by their respective branches of law rather than any unified category. 53 This reasoning was adopted in ByBit to support the idea that the category is not closed. 54 Low similarly argues that the category extends beyond

WS Holdsworth, ‘The History of the Treatment of Choses in Action by the Common Law’ (1920) 33 Harvard Law Review 997, 1001. 49 Law Commission (n 9) para 3.32. 50 Holdsworth (n 48) 1001 and 1010. 51 Holdsworth (n 48) 998. 52 Ibid. 53 Ibid. 54 ByBit (n 1) paras [34]-[35].

108

Singapore Comparative Law Review

Chaos In Action

enforceable rights. 55 He proposes a sub-classificatory system distinguishing different kinds of choses in action (e.g. contractual rights and intellectual property rights). Australia has embraced this wider conception. In National Trustees Executors and Agency Company of Australasia Limited v Federal Commissioner of Taxation (Cain’s Case),56 Kitto J held that choses in action are not limited to rights enforceable by a court, illustrating the point by reference to statute-barred debts and government bonds, which could not be the subject of a suit against the Crown in England until the Crown Proceedings Act 1947. Accordingly, Australian law requires no identified or identifiable counterparty. Jackman J similarly argues that creating a statutory third category would stultify the development of common law. 57 Australia has clearly embraced the wide conception. We disagree on two grounds. First, the wide conception dilutes the defining characteristics of choses in action in the narrow sense, which presently can still be clearly identified amongst all current choses in action within the UK and Singapore.58 As discussed in Section 4.2, cryptoassets, unlike choses in action, exist independently of whether any rights or claims relating to them are enforceable by action against an identifiable counterparty, 59 pursuant to AA.60 The UK and Singapore have not, ByBit apart, adopted Australia’s residual conception. The UK Jurisdiction Taskforce’s Legal Statement has established that there are some examples in case law of intangible property that are not choses in action (e.g. milk quotas, carbon emissions allowance, waste management licences). 61 They add that a number of important foundational statutes such as the Theft Act 1968 s 4, the Proceeds of Crime Act 2002 s 340, and the Fraud Act 2006 s 5 distinguish choses in action from other sorts of intangible property. This shows that the category of choses in action still has principled boundaries and is not so broad as to encompass all residual things that are not choses in possession. 55

Low and Hara (n 8); Kelvin Low, ‘Cryptoassets and the Renaissance of the Tertium Quid?’ in Chris Bevan (ed), Research Handbook on Property, Law and Theory (Edward Elgar 2024) 463. 56 (1954) 91 CLR 540 (HCA). 57 Ian Jackman, “What Has Taxonomy Ever Done for Us? UNCITRAL’s 2023 ‘Taxonomy of Legal Issues Related to the Digital Economy’” (2024 United Nations Day Lecture, UNCITRAL National Coordinating Committee for Australia, 24 October 2024) <https://www.fedcourt.gov.au/digital-law-library/judges-speeches/justicejackman/jackman-j-20241024> accessed 3 July 2026. 58 Law Commission (n 9) para 3.36. 59 Special Public Bill Committee, House of Lords (n 30), Q21 (Akber Datoo). 60 AA (n 14) at [55] (Bryan J). This principle is also recognised in Ruscoe v Cryptopia Ltd (in Liquidation) [2020] 2 NZLR 809 at [123] (Gendall J); Re Gatecoin Ltd (in Liquidation) [2023] 3 HKC 401 at [47] (Linda Chan J). 61 UK Jurisdiction Taskforce, Legal Statement on Cryptoassets and Smart Contracts (LawtechUK, November 2019) paras 82–83

109

Singapore Comparative Law Review

Chaos In Action

Second, in the wide conception, choses in action would experience different legal treatment. Treating cryptoassets as choses in action would limit the common law’s flexibility in developing new principles for this substantial group of property. This is especially important concerning areas like transfers, security and relative title. 62 Holdsworth himself notes that some rules devised for choses in action are "obviously inapplicable" to many incorporeal proprietary rights. 63 Yet those rules nevertheless continue to apply because the assets bear the same classificatory label unless "some authority, legislative or otherwise" provides otherwise. 64 Holdsworth regarded this confusion as the very affliction of the law of choses in action, which remains “confused, inconvenient and uncertain”. 65 Given this, ByBit’s reading of Holdsworth seems to have treated the wide conception with greater optimism than Holdsworth himself had intended. If a sub-classification were to be made within the category of choses in action, this already seems tantamount to creating a third category, which would be cleaner and more direct. The resulting doctrinal convolution is already apparent in cases like Armstrong DLW GmbH v Winnington Networks Ltd,66 where the court held that EUAs were neither choses in possession nor choses in action “in the narrow sense” — yet they, by Holdsworth’s definition, would be a chose in action in the wider sense. Critique of the case as having an “unduly convoluted complexion” arose from the judgment’s muddled stance on whether a narrow or wide conception was preferable, where the court was unable to analyse in a detailed way how legal title in an EUA would pass for the proprietary restitutionary claim that had been brought by Armstrong.67 It is unclear how classifying such assets as choses in action would offer greater legal clarity than a third category, when the result is the same taxonomy nested beneath a less accurate label. We develop the case for the legal consistency and clarity of a third category in Section IV. The boundaries of what constitutes a chose in action should remain drawn as per the narrow conception. Unduly expanding it to accommodate cryptoassets would burden a principled category with an unnecessarily byzantine complexion.

Law Commission (n 9) para 3.37. Holdsworth (n 48) 1029. 64 Ibid. 65 Holdsworth (n 48) 999. 66 [2012] EWHC 10 (Ch), [2013] Ch 156. 67 Kelvin FK Low and Jolene Lin, ‘Carbon Credits as EU Like It: Property, Immunity, TragiCO2medy?’ (2015) 27 Journal of Environmental Law 377, 377. 63

110

Singapore Comparative Law Review

III.

Chaos In Action

PROPRIETARY STATUS WITHOUT A SETTLED TAXONOMY — SINGAPORE’S CRYPTOASSET CASES AND THE P(DA)A 2025 A. Deliberate statutory minimalism in the P(DA)A 2025

The preceding discussion demonstrated why forcing cryptoassets into the orthodox categories carries substantial doctrinal costs. The English response has been to reject the exclusivity of the traditional binary through a deliberately minimalist approach with consequences left largely to the courts. Before the enactment of the P(DA)A 2025, the common law had begun to reject the assumption that the orthodox binary was exhaustive. Fry LJ’s dictum in Colonial Bank, once thought authoritative, was increasingly distinguished on its narrow facts, which turned on the interpretation of the bankruptcy statutes then in force. Professor Fox and Gullifer argue that it has been used as “authority for a proposition that it [was] not meant to support”.68 The UK Jurisdiction Task Force likewise emphasised legal flexibility, urging the common law to stretch “traditional definitions and concepts to adapt to new business practices”. 69 These judicial developments ultimately found statutory footing in the P(DA)A 2025 s 1, which implements the Law Commission’s recommendation by confirming that an object will not lose its status as an object of personal property merely because it falls outside existing categories. Importantly, the P(DA)A 2025 stops short of formally recognising a third category and places doctrinal development in the hands of the courts. Finally, the wider objective of the legislation has been described as positioning the UK as “the preeminent jurisdiction for the transaction of digital assets and the resolution of disputes arising from them”.70 Recognition of cryptoassets as personal property is a vehicle to bring much-needed stability to England, Wales and Northern Ireland at a significant juncture. As Datoo notes, the statute’s minimalism allows the “common law [to] add flesh to the bone”. 71 We also agree with Jackman J’s further argumentthat the common law is “one of the finest practical expressions of

D Fox and L Gullifer, Law Commission, Digital Assets: Call for Evidence – Responses (2022) 21. UK Jurisdiction Taskforce (n 15) para 77. 70 HL Deb 6 November 2024, vol 841, col 264GC (Lord Ponsonby of Shulbrede). 71 House of Lords (n 30) Q25 (statement by Akber Datoo). 69

111

Singapore Comparative Law Review

Chaos In Action

British empiricism” and it embodies the expectations of ordinary and honest people, particularly in a rapidly developing digital world.72 We will however caveat in Section 3.4 that the P(DA)A 2025 is much too bare because it lacks criteria for identifying third-category assets. Section IV proposes the statutory guidance we believe Singapore should provide if it were to adopt a similar Act. B. ByBit, Quoine, CLM v CLN And Janesh As Recognition Of Proprietary Status Without Classification In Singapore, case law on property rights in relation to cryptoassets has been piecemeal. At first instance, Quoine Pte Ltd v B2C2 Ltd accepted that it could constitute trust property, while leaving open the precise nature of that proprietary right. 73 The Court of Appeal, however, expressly declined to determine the issue, holding that it was unnecessary to decide whether Bitcoin could be held on trust because the alleged trust failed for lack of certainty of intention. 74 The proprietary status of cryptoassets therefore remained an open question. This was followed by CLM v CLN in 2022, which held that cryptocurrency could give rise to a proprietary right protected via an injunction.75 Janesh then held—albeit on an ex parte application—that NFTs satisfied the Ainsworth criteria for property. In 2023, ByBit classified USDT as a chose in action. In VOW v VOV, the court likewise assumed, without analysis of their proprietary status, that the wife’s cryptocurrency investments formed part of the matrimonial asset pool and were therefore divisible.76 With the UK having recognised the possibility of a third category, the question is whether Singapore should consolidate its current crypto-friendly approach in the same way. Not only are the current cases part of a piecemeal development, they are also limited by their procedural contexts. Janesh and CLM concerned interlocutory applications. The court in Janesh stressed that its task was simply to determine whether there was a serious question to be tried and was decided urgently ex parte, without submissions from the defendant. 72

Ian Jackman, ‘Is Cryptocurrency Property?’ (Commercial Law Association, 21 June 2024) <https://www.fedcourt.gov.au/digital-law-library/judges-speeches/justice-jackman/jackman-j-20240621> accessed 4 July 2026. 73 [2019] SGHC(I) 3 74 [2020] SGCA(I) 2 [144]. 75 [2022] SGHC 46. 76 [2023] SGHCF 9 [10].

112

Singapore Comparative Law Review

Chaos In Action

Thus the law remained open and a different conclusion could have been made given fuller submissions.77 These cases must be read in the context that the judiciary was not afforded an opportunity for fully contested consideration of cryptoasset proprietary classification. C. Deliberate minimalism in the UK and incidental drift in Singapore The common aversion to taxonomic classification masks materially different reasons for their reticence. The UK has deliberately adopted statutory minimalism, delegating the classificatory role to courts after a lengthy Law Commission process. In contrast, Singapore’s treatment of cryptoassets emerged from a sequence of urgent and interlocutory decisions and has not been ratified by any appellate court or the legislature. Singapore’s approach, however careful and considered in its discrete judgments, displays an incidental drift toward its current position on cryptoassets as a whole. A legislative decision to leave the incidents of a novel category to the common law can be supplemented, defended and ultimately amended. The significance of this choice is illustrated by the debate over the Law Commission’s view that statute is too rigid and that the treatment of cryptoassets should develop through responding to how they function in practice. 78 Consultees, such as the Aberdeen joint working group, pressed the opposing viewpoint that the undefined category risks becoming a vague miscellany, and Sean Edwards proposed that the Commission's criteria be enacted as tramlines for predictable development.79 Although the proposals were declined, the openness of the UK’s approach laid the path for contestation and documented resolution. The Singapore High Court’s treatment of USDT as a chose in action in ByBit has undergone no equivalent scrutiny. It arose from a summary judgment application where cryptoasset taxonomy was neither tested on appeal nor examined by Parliament. The doctrinal costs incurred by Singapore’s jurisprudence, as developed in Section 2.3, are byproducts of litigation, whereas a comprehensive reform project modelled on the UK’s approach could circumvent that strain or, at the least, provide a platform on which those costs may be openly acknowledged and reconciled through collective deliberation.

Janesh (n 24), [69]. Law Commission, Digital Assets as Personal Property: Supplemental Report and Draft Bill (Law Com No 416, HC 188, 2024) paras 3.79, 3.83. 79 Ibid, paras 3.20-3.21. 78

113

Singapore Comparative Law Review

Chaos In Action

D. Remedial uncertainty and institutional limits: Ping Fai Yuen v Fun Yung Li Section 3.1 argued that despite the P(DA)A 2025 confirming that the orthodox binary is nonexhaustive, it remains overly minimalist. Ping, decided within months of the Act’s commencement, demonstrates the consequences of leaving the common law to develop the doctrine with little statutory guidance. 80 Ping was the first English judgment applying the P(DA)A 2025. It shows that the UK’s bare approach of recognising property outside the orthodox binary does not assist adequately in resolving every doctrinal consequence flowing from that recognition. The court in Ping remained bound by OBG, which, although formulated in a period preceding the emergence of cryptoassets, remained “a clear block” to extending conversion to intangible assets such as Bitcoin. 81 Citing Lord Walker’s observation in OBG that reshaping this area is a matter for Parliament, the court struck out the conversion claim. Ping reveals two hurdles posed by the UK’s statutory regime. First, courts may lack the institutional expertise of the Law Commission or Parliament to resolve novel cryptoasset issues. Second, this creates an institutional impasse in which Parliament looks to the common law to elaborate the consequences of statutory recognition of a third category, while the courts, wary of enlarging established proprietary remedies, return the question to the legislature. Notably, we are left with questions as to whether statute should set out bespoke remedies for cryptoassets, such as insolvency and debt enforcement (e.g. rules concerning good faith acquisition). The UK’s priority of encouraging cryptoasset markets to flourish is jeopardised by the trade-off in legal certainty caused by these hurdles. It is precisely in such a rapidly developing market that investors require predictable legal outcomes. If uncertainty surrounding litigation persists, the resulting chilling effect may cause the P(DA)A 2025 to remain nothing more than a useful first step, with not much to follow.

80 81

Ping (n 7). Ibid, [66].

114

Singapore Comparative Law Review

Chaos In Action

This concern is particularly acute in Singapore, where the economy’s backbone is small- and medium-sized enterprises (“SMEs”) which employ 70% of the workforce.82 Unfortunately, for them, ‘costly, complex and time-consuming legal disputes” sometimes result in business closure.83 Litigation aside, they may instead resolve disputes through arbitration or commercial settlement. As such, there may be a lacuna left in appellate clarification on which incremental development depends. In addition, it presents the risk of continued piecemeal development. As industry consultees observed, litigation is fact-specific, limiting the wider applicability of individual decisions.84 They rightly caution that common law development cannot by itself provide the comprehensive guidance that market participants require. In a field developing as rapidly as cryptoassets, the pace of litigation may not keep pace with the needs of commerce. Section IV therefore proposes a novel set of statutory indicia that supply the initial classificatory framework. Parliament need not prescribe every remedy or determinatively ascertain the outcome for every asset but is certainly better placed to settle the threshold policy choices governing recognition. E. Following, tracing, and the persistence of cryptoasset identity through transfer Recovery may also fail at the anterior stage of establishing whether a cryptoasset retains its identity through transfer, that is, whether it remains the same res after on-chain transfer or if the transaction creates a substitute asset. ‘Following’ locates the original asset where it has not changed its form; ‘tracing’ identifies a substitute linked to the misappropriated asset. 85 The persistence of a cryptoasset’s identity determines the viability of following and tracing, which are distinct evidential processes, each identifying property against which a proprietary claim can be made.

Eric Lim, ‘SG60: How Singapore’s SMEs Are Shaping a Sustainable Future for Asean’ The Straits Times (13 August 2025) <https://www.straitstimes.com/singapore/sg60-how-singapores-smes-are-shaping-a-sustainablefuture-for-asean> accessed 4 July 2026. 83 OECD, Supporting Businesses through Better Justice Systems: A Focus on SMEs and Entrepreneurship (OECD Public Governance Policy Papers No 79, OECD Publishing 2025), 3 <https://doi.org/10.1787/1791ca66-en> accessed 11 July 2026. 84 Ministry of Justice, Digital Assets: Responses to Consultation (December 2025), 7 <https://cdn.websitebuilder.service.justice.gov.uk/uploads/sites/54/2025/12/Digital-assets-collated-consultationresponses.pdf> accessed 4 July 2026. 85 Foskett (n 26) 127 (Lord Millett).

115

Singapore Comparative Law Review

Chaos In Action

The choice between the competing legal characterisations of transfer-induced state changes is a normative question mistakenly understood as a factual one. 86 The continuity of the cryptoasset’s legal identity is ascribed by law, notwithstanding the physical observation that the post-transfer output may be a different data object than the pre-transfer one. For that reason, under Section 4.2, although persistence of identity is one of our supplementary indicia, it cannot itself determine whether an asset belongs in the proposed third category; it may only assist the court in resolving borderline cases. On an extinction/creation analysis, the pre-transfer object of personal property rights is extinguished and a new one is created in its place, such that following is inapplicable and the claimant is confined to tracing. On the persistent thing analysis, the object of personal property rights survives transfer. This article agrees with the Law Commission that there is no single correct answer, noting the consultees’ preference for the view that “a notional quantity unit can persist through a transaction”, and that no state change should be presumed to generate a presumption in favour of following or tracing, the applicable process being settled case by case with reference to the technology.87 As a cryptoasset system may implement a technically valid instruction regardless of the legal authority of the individual giving it, an on-chain state change cannot by itself establish the passing of legal title. Adapting Webb's economic-reality test, Chan locates persistence in the economic value of the transactional ability, explaining why a post-transfer output that is “economically and qualitatively identical” to its predecessor is treated as the same thing. 88 For chattels, this was resolved through specificatio—the doctrine by which a thing worked into a new form is treated as a new thing only if it’s essential identity has not survived. Intangibles escaped similar scrutiny because they “simply cannot be factually transferred without a concomitant conveyance of legal title”.89 Identity is ascribed by reference to an asset’s essential features, but cryptoassets’ idiosyncratic architecture means those features vary. As illustrated in Stephen Wilden v Person Unknown and Huobi Global S.A., following was held available in principle because expert evidence showed that identity was preserved through mixing, a conclusion that turned on the evidence more so than any settled ontology of Bitcoin.90 As a precondition for the same feature-specific approach to work for 86

Chan (n 5), 482. Law Commission (n 9), para 2.76, 9.30, 9.32, 9.34. 88 Chan (n 5), 494–95. 89 Ibid, 480, 497. 90 [2026] EWHC 1355 (KB) [40]. 87

116

Singapore Comparative Law Review

Chaos In Action

non-fungible tokens, their on-chain identity (e.g. contract address and token identifier) must be kept distinct from the off-chain artwork, metadata or rights they represent. The absence of a universal rule does not preclude a single framework from engaging with both fungible and nonfungible tokens where identity is fixed holistically. Whichever analysis applies, the ordinary rules on derivative transfer of title still govern. Persistence of identity as a standalone characteristic cannot determine if an asset belongs in the third category since an ordinary chose in action may also retain its identity upon assignment. It therefore operates only as a supplementary indicium in borderline cases. IV.

A PROPOSED FRAMEWORK FOR THE IDENTIFICATION OF THIRD CATEGORY PERSONAL PROPERTY A. Layer 1: four core presumptive indicia

We identify four core criteria which, taken together, are intended to function as the primary classificatory test for the proposed third category. These are drawn from academic literature and case law, and are a cluster of characteristics which recur whenever courts find sufficient reason to treat a cryptoasset as property. Where an asset satisfies all four, it should be treated as falling within that category. Where it does not, the analysis proceeds to a second layer, comprising factors relevant to judicial discretion in borderline cases. We propose a structured framework that is deliberately non-exhaustive: the first layer identifies the paradigm case, while the second accommodates technological and doctrinal variation at the margins. Given this, the first layer concerns the assets’ intrinsic characteristics and the second layer indicia examines how the assets behave in practice. We propose that these are adopted into legislation or statutory guidance by Parliament. (a) Rivalrousness Rivalrousness is the most conceptually distinct of our four proposed indicia, directly bearing on why cryptoassets resist classification as choses in action and identified by the Law Commission as “fundamentally important”. 91

Law Commission (n 9), paras 4.30-4.32.

117

Singapore Comparative Law Review

Chaos In Action

This article adopts the definition of rivalrous set out in the Digital Assets (Scotland) Act 2026: 92 An electronic system makes a thing rivalrous if— (a) the system maintains an immutable record of transactions in relation to the thing, and (b) that record is used to ensure that when, within the system, a person transacts in relation to the thing in a certain way (for example by transferring or spending it), the person loses the ability to transact in relation to the thing in that way again. The indicium of rivalrousness serves a useful purpose in distinguishing cryptoassets from pure information. Pure information can usually be replicated without limits, and one cannot control pure information in a meaningful way that excludes simultaneous use by others. Meanwhile, in Ruscoe v Cryptopia Ltd, a New Zealand case, it was held that cryptoassets were designed to be items of tradable value, not merely to convey information. 93 This has led to mechanisms being innovated such that cryptoassets cannot be double-spent (as explained in Tulip Trading Ltd v Bitcoin Association for BSV).94 This includes “robust technical authentication and validation mechanisms which ensure that the same crypto-token cannot be consumed twice or associated with two public addresses”.95 Although the data that a cryptoasset encompasses can indeed be copied through forking, each cryptoasset has a unique purpose within the system which cannot be duplicated. Consequently, a cryptoasset has a “form and function” that together make it rivalrous.96 (b) Independence from the legal system The second indicator is a measure of independence from the legal system. An asset satisfies this criterion if it has a factual existence that the legal system must accommodate but does not create. No legal rule can create or destroy a cryptoasset independent of the underlying facts; no court decision can effectively say that it has ceased to exist. This is supported by the Law Commission’s Final Report, where cryptoassets were said to “exist independently of the rights and claims 92

Digital Assets (Scotland) Act 2026 (asp 12), s 1(2). Ruscoe (n 60). 94 [2023] EWCA Civ 83 [24]. 95 Law Commission (n 9), para 4.33. 96 Law Commission (n 9), para 4.45. 93

118

Singapore Comparative Law Review

Chaos In Action

associated with them and are used and enjoyed independently of whether they give rise to rights enforceable by action”. 97 Indicium (b) is most helpfully illustrated via its contrast with conventional choses in action. A debt may cease to exist if a court declares the underlying obligation settled but cryptoassets e.g. bitcoin do not. Simply put, this characteristic excludes bank balances, debts and other choses in action whose existence is entirely contingent on legal enforceability. (c) Control through cryptographic mechanism The third criterion is distinct from, and should not be conflated with, criterion (b). Where (b) asks whether the asset needs the law to exist, (c) asks whether the holder needs the law to control it. It concerns the mechanism by which a holder exercises exclusive dominion over the asset rather than the asset's relationship to the legal order as a whole. For cryptoassets, control and exclusivity are exercised by the cryptographic authentication process which allows only the holder of the private key to deal with the asset, such as in its transfer or exchange, to the exclusion of others. 98 This technological reliance for ownership is foreign to both of the orthodox categories. This was aptly summarised in Farnhill DHCJ’s judgment in D'Aloia v Persons Unknown which held that USDT, like Bitcoin, was built on “a system that replaced the need for trust with cryptography rather than rights”. 99 (d) Absence of a counterparty obligor Ownership of a decentralised cryptoasset does not, merely by virtue of that ownership, confer a claim-right against any identified issuer or debtor. This is another characteristic distinguishing cryptoassets from choses in action in the orthodox sense and explains why they cannot be accommodated within the category without doing violence to its established boundaries. As Timothy Chan has observed, a holder of Bitcoin "enters into no bilateral contract with any particular counterparty" and individual nodes "do not expect to undertake liability to cryptoasset

Law Commission (n 9) para 3.68. Kate Allass, ‘Cryptoassets: A Guide on Key Legal Issues’ (Farrer & Co, 12 February 2020) <https://www.farrer.co.uk/news-and-insights/cryptoassets-a-guide-on-key-legal-issues2/> accessed 4 July 2026. 99 [2024] EWHC 2342 (Ch) [75]. 98

119

Singapore Comparative Law Review

Chaos In Action

holders in the course of mining" — such that there is, in substance, no identified person against whom ownership of the asset alone generates a claim.100 This criterion also operates as a negative classifier. Where a cryptoasset derives its value from a promise or obligation owed by an identifiable issuer, as with a stablecoin redeemable for fiat currency or a tokenised bond carrying repayment obligations, the asset begins to resemble a traditional chose in action, because its worth depends on a legal relationship with a counterparty rather than on its independent existence on the blockchain. B. Layer 2: two supplementary indicia for ambiguous cases The first layer is treated as determinative for the purposes of this framework in the paradigm case. The second is residual, and applies only where the asset does not fully satisfy the first-layer criteria. It provides two further indicators that may assist in determining whether the asset is nevertheless sufficiently analogous to the paradigm case. (a) Susceptibility to unauthorised factual interference An asset that can be transferred, dissipated, or wrongfully interfered with through unauthorised access to the relevant system exhibits a feature that property law is designed to address. Cryptoassets more often exhibit wrongful interference in a form that is direct and not reducible to breach by an identified obligor. Unlike a chose in action, which is ordinarily vindicated through enforcement against an identified obligor, a cryptoasset may be dealt with through factual control within the system itself. The fact that cryptoasset disputes so often arise out of hacking, fraud, and unauthorised transfers is therefore significant. 101 Again, we find it helpful to look at a negative example, where an asset cannot be dealt with in this way without the intervention or complicity of a platform operator, that dependence suggests that it more closely resembles a platform-based entitlement than a distinct third category asset.

100 101

Chan (n 5), 493. Law Commission (n 9) para 9.42.

120

Singapore Comparative Law Review

Chaos In Action

(b) Persistence of identity through transfer A further indicator is whether the asset retains a distinct and traceable identity as it moves between parties. Where it does, this supports the view that the asset is being dealt with as an object in its own right, distinct from any claim against another person. Farnhill DHCJ’s analysis is consistent with this approach, treating USDT as capable of remaining identifiable through successive transfers for the purposes of tracing and following. 102 C. Application of the proposed framework Table 1: Classification of representative cryptoassets under the proposed framework Asset

Classification

Principal reason

Bitcoin/ Ether

Within third category

Meets all four Layer 1 criteria.

NFTs

Within third category

Meets all four Layer 1 criteria.

Stablecoins (e.g. USDT)

Likely

within

the

third On-chain control supports third-

category, but requires Layer category 2 analysis.103

treatment,

but

issuer

obligations which engage indicium (d) may point towards a chose in action.

Memecoins/ altcoins

Requires Layer 2 analysis

Classification

depends

the

system’s governance and control structure.

102

D'Aloia (n 99) [7(ii)]. The D'Aloia (n 99) judgment definitively classifies USDT as a distinct form of property outside the orthodox binary at [5] though it is worth noting that this decision comes before P(DA)A 2025, it nonetheless diametrically opposes the conclusion reached on the same asset in ByBit. 103

121

Singapore Comparative Law Review

Digital carbon credits

Chaos In Action

Likely within third category

Designed to be rivalrous and not dependent on a single issuer’s obligation.

In-game digital assets 104

Outside third category

Dependent on the platform operator for existence and control.

Electronically issued debt Chose in action

Depends

enforceable

securities

obligation owed by an issuer.

This table summarises the application of the proposed framework but does not purport to state the current law in either jurisdiction. Assets classified as requiring Layer 2 analysis depend on the technical architecture and legal structure of the system in question. D. Limitations with our model This section examines three prevailing grounds for non-adoption and argues that none warrant rejection of the model. Technological Neutrality and Undue Expansion of Categories We begin by considering two related objections. The first is that a third category is insufficiently technologically neutral. The second is that a proliferation of categories follows as its corollary. A category tailored to the characteristics of cryptoassets is not technologically-neutral and is said to rest on a factual substrate that may not survive. Recent research suggests that the timeline for quantum attacks on the elliptic-curve cryptography underlying major blockchains is shortening, exposing transactions to possible attacks while they await confirmation. 105 For example, an

104

In R v Lakeman [2026] EWCA Crim 4, the court held that in-game gold pieces fell within the Theft Act 1968 definition of property. We would like to note that the statutory criminal definition differs from the private law category, which we explore in this essay. 105 Craig Gidney, ‘Quantum Computing Attack Timeline to Consider when Migrating to Post-Quantum Cryptography’ (31 March 2026) arXiv <https://arxiv.org/abs/2603.28846> accessed 4 July 2026.

122

Singapore Comparative Law Review

Chaos In Action

attacker able to derive a private key from a public key could authorise transactions indistinguishable from those of the former controller, causing our third indicium (control through cryptographic mechanism) to become unstable. In spite of any failure of the hash mechanism, the tokens “might still be rivalrous in that they could not be double-spent”, while “their factual excludability could be degraded to such an extent that it was no longer possible to exclude others from the use or enjoyment of those tokens”.106 Proposed short-term initiatives such as the Bitcoin Improvement Proposal 360 and the SPHINCS+ / SLH-DSA signature schemes contemplate a mass technological transition and could mitigate this risk.107 Quantum vulnerability is accordingly not necessarily fatal to cryptoassets as a class, although it does not eliminate the legal concern as migration is technically, economically and institutionally difficult. The modest conclusion is that cryptoassets are unlikely to disappear entirely but may evolve with the factual substrate of our indicia shifting within the short term. Courts and Parliament must be prepared to accept that an asset can move into or out of the category, which should be recognised in the statute. A cryptoasset that initially satisfies the four indicia but is later fundamentally compromised (e.g. permits double-spending and thus ceases to be rivalrous) should be capable of losing its third-category status. It might be objected that this undermines legal certainty for courts and commercial parties. This article contends instead that stability lies in the legal test rather than the permanence of any asset's status. The same legal criteria apply throughout; what evolves is the factual substrate to which those indicia are applied. The second objection is that recognition would encourage unnecessary proliferation of property categories, muddling the idea of personal property and the original distinction between choses in action and choses in possession. This raises a further doctrinal objection, that creating a third category without defined boundaries or incidents may replace one ambiguity (of whether the asset is property) with several others (concerning what rules govern transfer, security, priority and liability). Both are overstated. The threshold to establish a new category is high, as the property must be distinguished from the three existing categories, which, as observed through the creation 106

Law Commission (n 9) para 4.33(2). Omkar Godbole, ‘Bitcoin’s $1.3 Trillion Security Race: Key Initiatives Aimed at Quantum-Proofing the World’s Largest Blockchain’ CoinDesk (5 April 2026) <https://www.coindesk.com/tech/2026/04/04/bitcoin-s-usd1-3trillion-security-race-key-initiatives-aimed-at-quantum-proofing-the-world-s-largest-blockchain> accessed 4 July 2026. 107

123

Singapore Comparative Law Review

Chaos In Action

of the P(DA)A 2025 is a long and arduous affair. As to the second concern, statutory recognition is a starting point. As such, the relevant question is whether the gap it leaves is smaller than the uncertainty produced by forcing cryptoassets into the existing dichotomy, which would distort the position of litigants and the business community. For the reasons given in Sections II and III, the uncertainty of a third category is more transparent and the proposed indicia enable courts to reach more principled and predictable conclusions. While the indicia do not eliminate borderline cases, novel technologies will always generate difficult questions. Our indicia redirect litigation to the application of articulated statutory criteria, not the anterior question of the existence of the category itself. That inquiry is narrower, capable of producing greater consistency without depriving the common law of its capacity to adapt. Limitations to do with Rivalrousness Watts and Low disagree with the Law Commission on the nature of rivalrousness in cryptoassets.108 The Law Commission locates rivalrousness as inherent to technology, whereas Watts and Low contend that it is conferred by law through the recognition of cryptoassets as property. More specifically, legal recognition of cryptoassets as property confers legal rivalrousness “upon the power relation between a subject and his cryptoassets”. 109 Further, they read the Commission’s discussion of “fragile rivalrousness” as implying that rivalrousness is not absolute.110 It is important to exercise caution in this regard. The Commission only accepts excludability as dynamic and thus “fragile”, explaining that the practical ability to exclude “is likely to exist as a graduated quality that manifests on a continuum”. 111 The second limb of the disagreement conflated two prongs of rivalrousness, taking the concession on excludability to be for rivalrousness in its entirety. Our indicia hold the two qualities apart. Indicium (a) is a binary evaluation of whether the protocol prevents a token from being spent twice, whereas indicium (c) is a matter of whether the holder may in fact keep others from the asset, a question of degree such that the ability to exclude may be eroded without any token thereby becoming capable of being spent twice.

108

Peter Watts and Kelvin FK Low, ‘The Case for Cryptoassets as Property’ in Sinéad Agnew and Sir Marcus Smith (eds), Law at the Cutting Edge: Essays in Honour of Sarah Worthington (Hart Publishing 2024) 281, 285–286. 109 Ibid, 288. 110 Law Commission (n 9) para 4.34. 111 Ibid, para 4.33(2).

124

Singapore Comparative Law Review

Chaos In Action

We prefer the Law Commission's approach on the first point, which better reflects how market participants interact with cryptoassets and has been adopted at the Court of Appeal level. Tulip Trading recognised that the software, operated by a network of participants, is the core mechanism that ensures they cannot be double-spent. Hence, crypto-tokens are, to use the Law Commission’s terminology, “mathematically scarce”. 112 This practical scarcity distinguishes them from other forms of property. The Law Commission’s recognition of the rivalrousness of cryptoassets being “fragile” acknowledges the risks inherent to digital technology. Moreover, while Watts and Low argue that rivalrousness arises from the operation of law, cryptoassets exist and function independently of legal recognition. Were a court to deny that they are property, they would remain unspendable twice. Were mechanisms like hashing and the prevention of double-spending to fail, they would cease to be rivalrous regardless of the law’s verdict. Rivalrousness is inherent in the design of the cryptoasset and evidently not conferred by legal recognition. Finally, Clover Alcolea and Mihal argue that 51% and 34% attacks by hackers target cryptoasset systems to permit double-spending, thereby undermining rivalrousness. 113 This point engages indicium (a) and is answered on the indicium’s own terms. Where an attack demonstrates that a system’s consensus mechanism has ceased to operate, the asset fails indicium (a) and is resolved, as the section above contemplates, by an asset exiting the category. V.

CONCLUSION

While Singapore’s engagement with the proprietary status of cryptoassets is welcome, such recognition has outpaced a coherent account of the kind of property being recognised. This had the consequence of arriving at choses in action as an answer via a single classificatory statement in ByBit, widening a principled category into a residual one, at the cost set out in Section 2.3. Therefore, this article proposes a two-layer structure of presumptive indicia to delineate the boundaries of the third category, which could be set out in legislation or legislative guidance and refined by the courts through the common law. The first layer comprises the core identifying

112 113

Law Commission (n 9) para 4.45. Clover Alcolea and Mihal (n 22), 53.

125

Singapore Comparative Law Review

Chaos In Action

indicia of rivalrousness, independence from the legal system, control through cryptographic mechanism and absence of a counterparty obligor. The second comprises the supplementary indicia of susceptibility to unauthorised factual interference and persistence of identity through transfer that resolve assets at the periphery. A court applying them would decide on the architecture of the asset before it instead of how far an inherited dichotomy may be stretched to contain the cryptoasset. The classification of personal property would then take a tripartite form, three distinct categories capable of orienting doctrine as the technology continues on its rapid and destabilising development. Legislative recognition of certain cryptoassets as a third category, leaving the specific incidents to the common law, would supply the clarity that incremental litigation cannot produce at the pace commerce requires. Finally, cryptoassets would be classified by what their architecture dictates, displacing the negative inference that anything incapable of possession must be a chose in action, and sparing the orthodox categories the chaos that indiscriminate expansion invites.

126

Shape your ambition into impact

Apply now and start your journey with us.

Our unique Hong Kong eight-seat training contract offers you a wide breadth of experience across our many areas of law.

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

SHOULD AI CHATBOTS BE LEGALLY PERMITTED TO GENERATE SEXUAL CONTENT FOR ADULTS’ “PRIVATE” GRATIFICATION — A JURISPRUDENTIAL ANALYSIS Michelle Yap ABSTRACT Many today posit that artificial intelligence (“AI”) chatbots should be legally permitted to generate adult sexual content for adults’ “private” gratification (i.e. adult sexual content that is only accessible by the one user interacting with the AI chatbot, without further dissemination or publication), especially because such “private” consumption “harms no one”. In this article, I leverage legal theory, philosophy and jurisprudence to analyse whether this position is sound. To begin, I define the key terms and scope of my research, and explain why the study of this question is timely and salient. Next, I identify six different types of such AI-generated adult sexual content for adults’ “private” gratification, and analyse how four jurisdictions – Australia, China, Singapore and the United Kingdom – regulate this question. Following this, I strive to apply four schools of jurisprudential theories – John Stuart Mill, the Hart-Devlin debate, the Rawls-Finnis debate and Virtue Ethics – to my research question, and analyse the relevant strengths and weaknesses of each theory. Considering this jurisprudential analysis and other practical factors, I conclude that AI chatbots should not be legally permitted to generate sexual content for adults’ “private” gratification, while recognising that no one-size-fits-all regulatory response should be necessitated as the details of regulation are jurisdiction-specific. Nevertheless, I note that such “private” gratification harms oneself by diminishing one’s freedom for excellence, that the morality of human actions does not rest on consequences alone but on one’s object, intention and circumstances, and echo the importance of upholding the inviolable dignity of every human person.



LLM (Cantab), LLB (SMU). The author is grateful to the editors of the Singapore Comparative Law Review, Joshua Wan and Natasha Wong, for their helpful comments. Any mistakes or omissions remain the author’s own. 1 Loosely speaking, as one could argue that “technological AI developments” and “the sound development of AI ethics, policies and regulations” are not entirely dichotomous.

127

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

INTRODUCTION1

1. In this Article, I hope to leverage legal theory, philosophy and jurisprudence to address the following question: “should artificial intelligence (“AI”) chatbots be legally permitted to generate sexual content for adults’ “private” gratification?” 2. This Article will proceed in five parts. First, some background on why the study of this question is timely and salient. Second, the scope and definition of key terms in my research. Third, an analysis of how different jurisdictions’ existing approaches to AI chatbot regulation apply to my specific research question at hand. Fourth, an application of jurisprudential theories to my research question. Fifth, an overarching recommendation on whether AI chatbots should be legally permitted to generate sexual content for adults’ “private” gratification, and other relevant factors for consideration. A. Why even ask this question? 3. The study of this question is timely and salient in view of recent global AI chatbot developments, which arguably suggest that technological AI developments are rapidly outpacing the sound development of AI ethics, policies and regulations. 1 For example, OpenAI may allow “verified adults” to use ChatGPT to generate erotic content through the launch of its “adult mode” in the future, although these plans have presently been put on hold;2 Meta’s leaked internal policy document revealed that Meta allows its AI chatbots to engage in “sensual” conversations with children;3 OpenAI’s latest public version of

Financial Times, “OpenAI puts erotic chatbot plans on hold ‘indefinitely’” (26 March 2026) <https://www.ft.com/content/de9bf0af-b241-424f-8229-5870b1c0d93d?syn-25a6b1a6=1> (accessed 13 June 2026); Anthony Ha, “OpenAI delays ChatGPT’s ‘adult mode’ again” (7 March 2026) <https://techcrunch.com/2026/03/07/openai-delays-chatgpts-adult-mode-again/> (accessed 13 June 2026); Ashley Belanger, “OpenAI’s own mental health experts unanimously opposed “naughty” ChatGPT launch” (17 March 2026) <https://arstechnica.com/tech-policy/2026/03/chatgpt-may-soon-become-sexy-suicide-coach-openai-advisorreportedly-warned/> (accessed 13 June 2026). 3 Jeff Horwitz, “Meta’s AI rules have let bots hold ‘sensual’ chats with kids, offer false medical info” (14 August 2025) <https://www.reuters.com/investigates/special-report/meta-ai-chatbot-guidelines/> (accessed 13 June 2026); Charlotte Edwards, “Meta investigated over AI having 'sensual' chats with children” (18 August 2025) <https://www.bbc.com/news/articles/c3dpmlvx1k2o> (accessed 13 June 2026); Jeff Horwitz, “Meta CEO Zuckerberg blocked curbs on sex-talking chatbots for minors, court filing alleges” (28 January 2026) <https://www.reuters.com/legal/government/meta-ceo-zuckerberg-blocked-curbs-sex-talking-chatbots-minors-courtfiling-2026-01-27/> (accessed 28 June 2026).

128

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

ChatGPT can “generate sexualised images… with a single prompt”; 4 and Elon Musk faced intense backlash for the recent “Grok undressing, nudification, and ‘bikinification’ trend on X”, where numerous users rampantly utilised the Grok AI chatbot to generate, publish and disseminate non-consensual sexualised and intimate images on X.5 While global AI and online safety regulators will always be confronted with the Collingridge dilemma, 6 this Article humbly seeks to encourage these regulators to rethink this question and the rationale behind this decision. B. Scope of research and definition of key terms 4. At the outset, I will define the scope and key terms in my research question. 5. For the scope, I will not be addressing the following areas of law as the topic of “AI chatbot regulation” is too vast — data protection law, intellectual property law, product liability and consumer protection law, defamation law, the tort of negligence, mental health-related 4

Chris Vallance, “ChatGPT can be made to generate sexualised and violent images, researchers find” (18 June 2026) <https://www.bbc.com/news/articles/c802ldjdklzo> (accessed 28 June 2026); Jim Nightingale, “ChatGPT Spontaneously Generates Sexual Violence and Hardcore Snuff Imagery” (16 June 2026) <https://mindgard.ai/blog/chatgpt-spontaneously-generated-violent-images-from-a-viral-prompt> (accessed 28 June 2026); 5 A.J. Vicens & Raphael Satter, “Elon Musk's Grok AI floods X with sexualized photos of women and minors” (4 January 2026) <https://www.reuters.com/legal/litigation/grok-says-safeguard-lapses-led-images-minors-minimalclothing-x-2026-01-02/> (accessed 13 June 2026); Federica Fedorczyk, “Expert Comment: Chatbot-driven sexual abuse? The Grok case is just the tip of the iceberg” (14 January 2026) <https://www.ox.ac.uk/news/2026-01-14expert-comment-chatbot-driven-sexual-abuse-grok-case-just-tip-iceberg> (accessed 13 June 2026); Ysa Marie, “Policing the Feed: AI-Generated Sexual Content on Social Media and Its Impacts on the Vulnerable” (25 February 2026) <https://rsis.edu.sg/rsis-publication/rsis/policing-the-feed-ai-generated-sexual-content-on-social-media-andits-impacts-on-the-vulnerable/> (accessed 13 June 2026); Lee Li Ying, “Deepfake nudes inflict real harm despite being fake: Experts” (9 February 2026) <https://www.straitstimes.com/singapore/deepfake-nudes-inflict-real-harmdespite-being-fake> (accessed 13 June 2026); European Data Protection Supervisor “Joint Statement on AIGenerated Imagery and the Protection of Privacy” (23 February 2026) <https://www.edpb.europa.eu/system/files/2026-02/2026.02.23-joint-statement-ai-generated-imagery-61signatories-distributed-21.02.2026.pdf> (accessed 13 June 2026); Elon Musk, Post on X <https://x.com/elonmusk/status/2011562747620712927> (accessed 28 June 2026); Justin Hendrix, “Breaking Down a Class Action Lawsuit Filed Over Grok 'Undressing' Controversy” (28 January 2026) <https://www.techpolicy.press/breaking-down-a-class-action-lawsuit-filed-over-grok-undressing-controversy/> (accessed 28 June 2026); Kali Hays, “Teens sue Musk's xAI over Grok's pornographic images of them” (17 March 2026) <https://www.bbc.com/news/articles/cgk2lzmm22eo?at_medium=RSS&at_campaign=rss> (accessed 28 June 2026). 6 Simon Chesterman, “From Ethics to Law: Why, When, and How to Regulate AI” (May 2023) NUS Law Working Paper No 2023/014 at pp 6-7; David Collingridge, The Social Control of Technology (Frances Pinter, 1980). As per Simon Chesterman, the Collingridge dilemma reveals how “any effort to control new technology faces a double bind. During the early stages, when control would be possible, not enough is known about the technology’s harmful social consequences to warrant slowing its development. By the time those consequences are apparent, however, control has become costly and slow”.

129

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

AI chatbot regulation (e.g. addictive design, suicide and self-harm risk, user wellbeing, etc), and AI transparency and accountability regulation (e.g. labelling and content provenance, record-keeping and risk assessments, training data transparency, etc). Instead, I will focus on online safety regulation, and specifically “sexual content” generated by “private AI chatbots” as defined in paragraph 6 below. 6. For the definition of key terms, I will accept “gratification” to refer to pleasure or satisfaction,7 “sexual content” to refer to multimodal digital material (e.g. images, videos, texts, audio) that describe, depict or promote sexual acts, nudity, or sexualised behaviour that are primarily intended for sexual arousal, 8 and “AI chatbots” to refer to nondeterministic generative AI systems, built upon large language models (“LLMs”), that enable users to generate content and are designed to simulate human-like relationships. 9 I will further sub-categorise “sexual content” into “adult sexual content” and “child sexual exploitation and abuse material” (“CSEAM”), as CSEAM is often recognised as a distinct category across numerous online safety regulations, and sub-categorise “AI chatbots” into “private AI chatbots”, 10 where the output from the AI chatbot(s) is only accessible by one user, and “non-private AI chatbots”,11 where the output from the AI chatbot(s) is accessible

Cambridge Dictionary, Definition of “gratification” <https://dictionary.cambridge.org/dictionary/english/gratification> (accessed 13 June 2026); Robert Nozick, Anarchy, State and Utopia (Blackwell, 1974). 8 However, content of literary, artistic, scientific or educational merit are not considered sexual content. Criminal Law of the People's Republic of China (1979); General Administration of Press and Publication Notice on Further Strengthening Management of Online Literary Publications (People’s Republic of China) (2020); Provisions on the Governance of the Online Information Content Ecosystem (People’s Republic of China) (2019); Online Safety Act 2023 (UK); Sexual Offences Act 2003 (UK); Obscene Publications Act 1959 (UK); Online Safety Act 2021 (Australia); Criminal Code Act 1995 (Australia); National Classification Scheme (Australia); Classification (Publications, Films and Computer Games) Act 1995 (Australia); eSafety Commissioner, Online Safety Codes and Standards Regulatory Guidance (April 2026) <https://www.esafety.gov.au/sites/default/files/202605/OnlineSafetyCodesandStandardsRegulatoryGuidance-April2026.pdf?v=1781353105087> (accessed 13 June 2026) (“eSafety Regulatory Guidance”); Broadcasting Act 1994 (Singapore); Penal Code 1871 (Singapore); Undesirable Publications Act 1967 (Singapore); Films Act 1981 (Singapore); Online Safety (Relief and Accountability) Act 2025 (Singapore). 9 eSafety Commissioner, Consolidated Industry Codes of Practice for the Online Industry (Class 1C and 2 Material) Head Terms at p 8 <https://onlinesafety.org.au/wp-content/uploads/2025/07/CLEAN-Phase-2-OSA-Codes-HeadTerms-1.pdf> (accessed 13 June 2026) (“eSafety Head Terms”); Soh Kian Peng and Vanessa Ng, “Chatbots and Liability for Negligent Misrepresentation” [2025] Singapore Academy of Law Practitioner 16; Quoine Pte Ltd v B2C2 Ltd [2020] SGCA(I) 2; Crime and Policing Act 2026 (UK) Amendment 433. 10 Using the term “private” loosely here. 11 Using the term “private” loosely here.

130

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

to more than one user(s).12 All of these “AI chatbots” may span across a range of functionalities including general-purpose, companion, enterprise, etc. 13 7. As this Article will focus on “private AI chatbots”, instead of “non-private AI chatbots”, I will further narrow my scope of “private AI chatbots” for study to: (i) standalone AI chatbots that are available via web browsers or apps (e.g. general-purpose ones such as ChatGPT, Gemini, Claude; companion ones such as Replika, Character.AI, Nomi.AI which may include avatars or characters); and (ii) embedded AI chatbots on social media services, messaging services, gaming services, etc (e.g. Grok on X, Meta AI on Facebook, Instagram and WhatsApp). However, I recognise that there are other unclassified AI chatbots such as user-created AI chatbots which may or may not be “private” (e.g. usercreated AI chatbots on Telegram, X, Instagram, Character.AI, Chub AI); enterprise, industry or sector-specific AI chatbots; other services which may or may not qualify as “AI chatbots”, let alone “private AI chatbots” (e.g. Sora); 14 and more AI chatbots that will likely be developed in the future.

Examples of “non-private AI chatbots” include ChatGPT group chats, Character.AI group chats and tagging @grok on X. Helena Horton, Dan Milmo & Amelia Gentleman, “Grok turns off image generator for most users after outcry over sexualised AI imagery” (9 January 2026) <https://www.theguardian.com/technology/2026/jan/09/grokimage-generator-outcry-sexualised-ai-imagery> (accessed 13 June 2026); OpenAI, “Introducing group chats in ChatGPT” (13 November 2025) <https://openai.com/index/group-chats-in-chatgpt/> (accessed 13 June 2026); Sarah Perez, “Character.AI introduces group chats where people and multiple AIs can talk to each other” (11 October 2023) <https://techcrunch.com/2023/10/11/character-ai-introduces-group-chats-where-people-and-multiple-ais-cantalk-to-each-other/> (accessed 13 June 2026); Ofcom, “Open letter to UK online service providers regarding Generative AI and chatbots” (8 November 2024) <https://www.ofcom.org.uk/online-safety/illegal-and-harmfulcontent/open-letter-to-uk-online-service-providers-regarding-generative-ai-and-chatbots> (accessed 13 June 2026) (“Ofcom’s open letter”). 13 Clare McGlynn et al, “Invisible No More: How AI Chatbots Are Reshaping Violence Against Women and Girls” (March 2026) at pp 34-35 <https://cronfa.swan.ac.uk/Record/cronfa71633> (accessed 13 June 2026); IBM, “Enterprise chatbots: Benefits and use cases” (13 January 2026) <https://www.ibm.com/think/topics/enterprisechatbot> (accessed 13 June 2026). 14 OpenAI, “Sora: Creating video from text” <https://openai.com/index/sora/> (accessed 13 June 2026); OpenAI, “What to know about the Sora discontinuation” <https://help.openai.com/en/articles/20001152-what-to-know-aboutthe-sora-discontinuation> (accessed 13 June 2026).

131

Singapore Comparative Law Review

II.

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

DO AUSTRALIA, CHINA, SINGAPORE AND THE UNITED KINGDOM

LEGALLY PERMIT AI CHATBOTS TO GENERATE SEXUAL CONTENT FOR ADULTS’ “PRIVATE” GRATIFICATION? 8. In this Article, I will analyse whether Australia, China, Singapore and the United Kingdom legally permit AI chatbots to generate sexual content for adults’ “private” gratification. As these four jurisdictions prohibit AI chatbots from generating CSEAM, 15 I will focus on adult sexual content, and specifically whether these four jurisdictions legally permit AI chatbots to generate the following six types of adult sexual content for adults’ “private” gratification (i.e. the adult sexual content is only accessible by the one user who prompted the AI chatbot, without further dissemination or publication): a. Non-consensual use of a real human person’s content to generate a deepfake, with an intention to disseminate or publish the deepfake; b. Non-consensual use of a real human person’s content to generate a deepfake, without an intention to disseminate or publish the deepfake; c. Fully AI-generated deepfake, with an intention to disseminate or publish the deepfake; d. Fully AI-generated deepfake, without an intention to disseminate or publish the deepfake;

Criminal Code Act 1995 (Australia); eSafety Commissioner, Phase 1 Unlawful Material Codes and Standards <https://www.esafety.gov.au/industry/codes/register-online-industry-codes-standards#register-of-industrystandards> (accessed 13 June 2026); eSafety Commissioner, “Frequently asked questions about access to online porn and other adult content” <https://www.esafety.gov.au/industry/codes/faq-access-to-online-porn-and-otheradult-content> (accessed 13 June 2026) (“eSafety FAQs”); Protection of Children Act 1978 (UK); Crime and Policing Act 2026 (UK); Penal Code 1871 (Singapore); Broadcasting Act 1994 (Singapore); Singapore, Code of Practice for Online Safety for Social Media Services <https://www.imda.gov.sg/-/media/imda/files/regulations-andlicensing/regulations/codes-of-practice/codes-of-practice-media/code-of-practice-for-online-safety-social-mediaservices.pdf> (accessed 13 June 2026) (“SMS Code”); Singapore, Code of Practice for Online Safety for App Distribution Services <https://www.imda.gov.sg/-/media/imda/files/regulations-and-licensing/regulations/codes-ofpractice/code-of-practice-app-distribution-services/code-of-practice-for-online-safety-app-distribution-services.pdf> (accessed 13 June 2026) (“ADS Code”); Criminal Law of the People's Republic of China (1979).

132

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

e. Fully AI-generated synthetic content, with an intention to disseminate or publish the synthetic content; and f. Fully AI-generated synthetic content, without an intention to disseminate or publish the synthetic content. 16 9. For the purposes of paragraphs 8a to 8f above, I clarify that AI-generated “sexual content” often manifests in four multimodal formats, or a mix thereof — images, videos, audios, texts. Additionally, I define “deepfakes” to refer to AI-generated content that realistically and reasonably depict another real human person, regardless of its source (i.e. nonconsensual use of another real human person’s content, or fully AI-generated), and define “synthetic content” to refer to AI-generated content that does not realistically or reasonably depict another real human person, and is fully AI-generated without the use of another real human person’s content. While these definitions are certainly not universal, I will adopt them for the ease of communication in this Article. 10. As the four jurisdictions directly or incidentally touch upon this question through a patchwork of different regulations, I will broadly analyse the following three buckets of laws: (i) criminal law, (ii) civil platform regulation, and (iii) remedies for victims. As this question focuses on “private” gratification, buckets (i) and (ii) will focus on ex ante obligations imposed on AI chatbot developers, deployers and users, 17 while bucket (iii) enhances contextual understanding, but is not completely relevant to my research question, because such remedies will require the victims to come forward. A. Australia 11. Australia legally permits AI chatbots to generate adult sexual content for adults’ “private” gratification, as explained below.

I will not cover “recklessness” for two reasons. First, to limit the scope of this Article. Second, all six types of sexual content are only accessible by the one user who prompted the AI chatbot, without further dissemination or publication. 17 For the purposes of this article, I will accept that AI chatbots have no distinct or separate legal personality, and hence will not delve into this extensive debate.

133

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

12. On criminal law broadly, Section 474.17A of Australia’s Criminal Code Act 1995, as amended by Schedule 1 of its Criminal Code Amendment (Deepfake Sexual Material) Act 2024, unlikely covers the adult sexual content in paragraphs 8a to 8f as it requires one to “transmit”18 the sexual material by “using a carrier service”, 19 and for the purposes of paragraphs 8e and 8f, “without consent”. Moreover, Australia’s Classification (Publications, Films and Computer Games) Act 1995 unlikely covers paragraphs 8a to 8f as it contains consistent references to the phrase “sell or deliver” for publications, 20 and “sale or exhibition” for films.21 13. On civil platform regulation, pursuant to Part 9 of Australia’s Online Safety Act 2021, 22 Australia has developed the Phase 1 “Unlawful Material Codes and Standards”, which cover Class 1A and 1B material, and the Phase 2 “Age-Restricted Material Codes”, which cover Class 1C and 2 material, as defined under Australia’s National Classification Scheme.23 As the Phase 2 Industry Codes cover AI chatbot-generated “online pornography that is inappropriate for children”, 24 while the Phase 1 Industry Codes and Standards only cover CSEAM,25 AI chatbot-generated adult sexual content are highly likely legally permissible for adults’ “private” gratification in Australia. This stance is reinforced by the fact that Australia has introduced age assurance requirements for certain standalone and

Criminal Code Act 1995 (Australia) s 474.17A(4). Id, at s 473.5; Telecommunications Act 1997 (Australia). 20 Classification (Publications, Films and Computer Games) Act 1995 (Australia) Part 3. 21 Id, at Part 2. 22 eSafety Commissioner, “Online Safety Codes and Standards” <https://www.esafety.gov.au/industry/codes#:~:text=Unlawful%20Material%20Codes%20and%20Standards,materi al%20and%20pro%2Dterror%20material> (accessed 13 June 2026) (“eSafety Codes and Standards”); eSafety Regulatory Guidance, supra n 8. 23 eSafety Regulatory Guidance, supra n 8, at p 6. 24 eSafety Codes and Standards, supra n 22; eSafety Commissioner, Phase 2 Designated Internet Services Online Safety Code (Class 1C and Class 2 Material) <https://onlinesafety.org.au/wpcontent/uploads/2025/07/Consolidated-Industry-Code-of-Practice-for-the-Online-Industry-Class-1C-and-Class-2Material-Schedule-6-%E2%80%93-Designated-Internet-Services-Online-Safety-Code-Class-1C-and-Class-2Material-1.pdf> (accessed 13 June 2026) (“eSafety Phase 2 DIS Code”); eSafety Commissioner, Phase 2 Social Media Services (Core Features) Online Safety Code (Class 1C and Class 2 Material) <https://onlinesafety.org.au/wp-content/uploads/2025/07/CLEAN-Phase-2-OSA-Codes-SMS-Code-Core-Features1.pdf> (accessed 13 June 2026) (“eSafety Phase 2 SMS Code”); Phase 2 Relevant Electronic Services Online Safety Code (Class 1C and Class 2 Material) <https://onlinesafety.org.au/wp-content/uploads/2025/07/CLEANPhase-2-OSA-Codes-RES-Code-1.pdf> (accessed 13 June 2026) (“eSafety Phase 2 RES Code”). 25 eSafety Regulatory Guidance, supra n 8. 19

134

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

embedded AI chatbots in its Phase 2 Age-Restricted Material Codes,26 and pursuant to the Online Safety Amendment (Social Media Minimum Age) Act 2024, although the scope of its definitions of an “AI companion chatbot feature” under the Phase 2 Head Terms, 27 and a “high impact generative AI DIS” under the Phase 2 Designated Internet Services Code, remains highly uncertain and debatable. 28 For example, does an “AI companion chatbot feature” cover Meta AI on Facebook and Instagram, or only Grok on X, given that Meta AI is arguably a “productivity-focused assistant” that is designed to “provide factual outputs” rather than “simulate personal relationships”?29 14. Moreover, Part 6 of Australia’s Online Safety Act 2021 does not cover the adult sexual content in paragraphs 8a to 8f as it requires “posting” or a “threat to post”, 30 and compliance with the baseline online safety expectations (“BOSE”) in Part 4 of the Online Safety Act 2021 is not legally enforceable31 — the BOSE mainly constitutes reporting obligations to enhance transparency, and only the failure to respond to such reporting obligations may be legally enforceable.32 Additionally, in response to FAQs on the Phase 2 “Age-Restricted Material Codes”, Australia’s eSafety Commissioner has explicitly stated that “[v]iewing pornography has long been legal in Australia but access to it is meant to be restricted to adults”. Lastly, the fact that X’s rules and policies explicitly allow for “adult content”, 33 and yet X continues to be permitted to operate for Australian end-users, strengthens my conclusion that Australia legally permits AI chatbots to generate adult sexual content for adults’ “private” gratification. 15. On remedies for victims, Australia’s Online Safety and Other Legislation Amendment (My Face, My Rights) Bill 2025 will amend the Online Safety Act 2021 and the Privacy Act 1988 to “establish a complaints and enforcement regime for the non-consensual sharing of 26

eSafety Phase 2 DIS Code, supra n 24; eSafety Phase 2 SMS Code, supra n 24; eSafety Phase 2 RES Code, supra n 24. 27 eSafety Head Terms, supra n 9; eSafety Phase 2 SMS Code, supra n 24; eSafety Phase 2 RES Code, supra n 24. 28 For example, the eSafety Phase 2 DIS Code may cover Grok on X but not Meta AI on Instagram and Facebook. 29 eSafety Head Terms, supra n 9. 30 Online Safety Act 2021 (Australia) s 75. 31 eSafety Regulatory Guidance, supra n 8, at p 32. 32 eSafety Commissioner, “Basic Online Safety Expectations” <https://www.esafety.gov.au/industry/basic-onlinesafety-expectations> (accessed 13 June 2026). 33 X Help Center, “Adult Content” (May 2024) <https://help.x.com/en/rules-and-policies/adult-content> (accessed 13 June 2026).

135

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

deepfake material” and “create a new statutory cause of action for the wrongful use or disclosure of deepfake material” respectively. 34 16. Overall, Australia legally permits AI chatbots to generate the adult sexual content in paragraphs 8a to 8f for adults’ “private” gratification. B. China 17. China legally prohibits AI chatbots from generating adult sexual content for adults’ “private” gratification, as explained below. 18. On criminal law broadly, Article 363 of the Criminal Law of the People's Republic of China criminalises the “production” of “obscene materials” 35 for the “purpose of profit”. However, it is evident from the Xuhui District People’s Court of Shanghai decision on Alien Chat,36 and China’s 200437 and 201038 Guidance, that the phrase “purpose of profit”

Online Safety and Other Legislation Amendment (My Face, My Rights) Bill 2025 (Australia). Criminal Law of the People's Republic of China (1979) Art 367. 36 While this case is not available online, and is pending appeal in the second-instance trial, reliable sources have reported on it. James Gong & Wang Yiting, “China TMT: Bi-monthly Update - January and February 2026 Issue” (3 April 2026) <https://www.twobirds.com/en/insights/2026/china/china-tmt-bi-monthly-update---january-andfebruary-2026-issue> (accessed 13 June 2026); China Law Translate, “Alien Chat: China’s 1st Case of Criminal Responsibility for AI-Generated Content” (20 January 2026) <https://www.chinalawtranslate.com/en/alienchat/> (accessed 13 June 2026) (“Alien Chat”); MLex, “China's first AI obscenity case tests criminal liability of companionship chat developers” (20 January 2026) <https://www.mlex.com/mlex/articles/2431692/china-s-first-aiobscenity-case-tests-criminal-liability-of-companionship-chat-developers> (accessed 13 June 2026); Sun Langchen, “China’s first AI companion app case enters second-stance trial, sparking debate on emotional AI service boundaries” (14 January 2026) <https://www.globaltimes.cn/page/202601/1353192.shtml> (accessed 13 June 2026); Mary Bennett & Rob Robinson, “When AI Becomes Accomplice: Shanghai Court Holds Developers Criminally Liable for Chatbot Content” (20 January 2026) <https://www.jdsupra.com/legalnews/when-ai-becomes-accompliceshanghai-3378572/> (accessed 13 June 2026); Zhang Yushuo, “China's First AI Pornography Case Is Adjourned at Second-Instance Trial” (15 January 2026) <https://www.yicaiglobal.com/news/chinas-first-ai-pornography-case-isadjourned-at-second-instance-trial> (accessed 13 June 2026); Xiao Xiao, “The ins and outs of the first "AI-assisted pornography case": AI chatted with users about pornography, and the platform was sentenced?” (15 January 2026) <https://m.21jingji.com/article/20260115/herald/945d8a99dba4a479057ff02a44776b3a.html> (accessed 13 June 2026). 37 People's Republic of China, Interpretation by the Supreme People's Court and the Supreme People's Procuratorate on Several Issues Concerning the Application of Law in Handling Criminal Cases Involving the Production, Reproduction, Publication, Sale, and Dissemination of Obscene Electronic Information Using the Internet, Mobile Communication Terminals, and Voice Information Platforms (30 September 2004) <https://www.cac.gov.cn/2004-09/30/c_126472347.htm> (accessed 13 June 2026) (“PRC’s 2004 Guidance”). 38 People's Republic of China, Interpretation (II) of the Supreme People's Court and the Supreme People's Procuratorate on Several Issues Concerning the Specific Application of Law in Handling Criminal Cases Involving the Production, Reproduction, Publication, Sale, and Dissemination of Obscene Electronic Information Using the Internet, Mobile Communication Terminals, and Voice Information Platforms (3 February 2010) <https://www.cac.gov.cn/2010-02/03/c_126472261.htm> (accessed 13 June 2026) (“PRC’s 2010 Guidance”). 35

136

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

is construed broadly such that China’s criminal law likely theoretically covers the multimodal39 adult sexual content in paragraphs 8a to 8f, although the factors taken into account for practical enforcement of Article 363 will likely be narrower. Where Article 363 is enforced against an AI chatbot developer(s), examples of such factors may include the volume of sexual content generated, the number of registered users and paying subscribers, the amount of profit gained, etc. 40 To illustrate, I will expound more on the Shanghai Court Alien Chat decision, although it is still pending appeal at the secondinstance trial.41 19. In Alien Chat, a Shanghai-based technology company leveraged foreign LLMs to create an AI companion app named “Alien Chat”, which allowed “registered users to create virtual characters or interact with characters created and shared by others”. 42 As this app generated a “large volume of content that graphically depicted sexual conduct or explicitly promoted pornography”, the Shanghai court cited Article 9 of the Interim Measures for the Management of Generative Artificial Intelligence Services 2023, noted that “Alien Chat was specifically designed and optimized to continuously generate pornographic content… [because its developers] exercised substantive control over outputs through the drafting and modification of system prompts”, and held the two developers of Alien Chat’s criminally liable pursuant to Article 363 of the Criminal Law. 43 20. There are three key unique factors in this Alien Chat decision. First, one-to-one “private” text-based chat logs between each user and the AI chatbot virtual character, which were not disseminated to third parties, constituted “obscene material” under China’s Criminal Law for the first time. 44 Second, the AI chatbot developers, instead of users, were held criminally liable.45 Third, although the prosecution sought the charge of “disseminating”

Criminal Law of the People's Republic of China (1979) Art 367. Alien Chat, supra n 36; MLex, supra n 36; Sun Langchen, supra n 36; Mary Bennett & Rob Robinson, supra n 36; Zhang Yushuo, supra n 36; Xiao Xiao, supra n 36; James Gong & Wang Yiting, supra n 36; PRC’s 2004 Guidance, supra n 37; PRC’s 2010 Guidance, supra n 38; China Law Translate, “Did China make sexting illegal?” (12 January 2026) <https://www.chinalawtranslate.com/did-china-make-sexting-illegal/> (accessed 13 June 2026). 41 James Gong & Wang Yiting, supra n 36. 42 MLex, supra n 36. 43 Ibid. 44 Xiao Xiao, supra n 36; MLex, supra n 36; Alien Chat, supra n 36. 45 Ibid. 40

137

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

obscene material for profit, the first-instance court changed the charge to “producing” obscene material for profit, thereby arguably calling into question whether the text-based chat logs, or the AI chatbot virtual avatar itself, constituted the “obscene material” in question.46 At this juncture, it is also notable that Article 80 of the Public Security Administration Punishments Law 2025 allows for imposition of detentions and fines on anyone who “produce[s]... obscene materials”. 21. On civil platform regulation, China generally prohibits AI chatbots from generating all types of sexual content in all multimodal formats, 47 also known as an “outright ban”, and hence does not impose any age assurance requirements. Examples of such regulations include the Interim Measures for the Management of Generative Artificial Intelligence Services 2023,48 the Provisions on the Governance of the Online Information Content Ecosystem 2019,49 the Provisions on the Administration of Deep Synthesis Internet Information Services 2022,50 the Provisions on the Management of Algorithmic Recommendations in Internet Information Services 2021,51 the Provisional Measures on the Administration of Human-like Interactive AI Services 2025, 52 and the Cybersecurity Law of the People's Republic of China. 53 22. On remedies for victims, Book Four of the Civil Code of the People’s Republic of China covers "Personality Rights”, including the “Rights to Likeness” 54 and “Rights to Privacy and Protection of Personal Information”. 55 23. Overall, China legally prohibits AI chatbots from generating adult sexual content in paragraphs 8a to 8f for adults’ “private” gratification.

Xiao Xiao, supra n 36; Alien Chat, supra n 36. I.e. images, videos, texts, audio. 48 Especially Articles 4, 9 and 14. 49 Especially Articles 6 and 7. 50 Especially Articles 6 and 4. 51 Especially Articles 9 and 4. 52 Especially Article 7. 53 Especially Article 13. 54 Civil Code of the People’s Republic of China (2020) Book Four, Chapter IV. 55 Ibid. 47

138

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

C. Singapore 24. Singapore legally permits AI chatbots to generate adult sexual content for adults’ “private” gratification, as explained below. 25. On criminal law broadly, the Criminal Law (Miscellaneous Amendments) Act 2025 amended, amongst other things, the Penal Code 1871 (“PC”). However, the amended Section 377BE of the PC likely only covers paragraph 8a and 8c at best, because of the phrases “threatens the distribution” and “without B’s consent”, 56 while the amended Section 377BD of the PC likely only covers paragraphs 8a to 8d at best, because of the phrase “of another person (B)”. 57 Furthermore, Section 377BE appears to only cover images and videos,58 although Section 377BD appears to cover images, videos, audios and texts.59 26. Moreover, although Sections 11 and 12 of the Undesirable Publications Act 1967 (“UPA”) may cover images of adult sexual content in paragraphs 8a to 8f, 60 they will unlikely cover such videos, texts and audios given the definitions of “publication”, “film” and “sound recording” in Section 2 of the UPA. Additionally, although Sections 29 and 30 of the Films Act 1981 (“FA”) may cover videos of adult sexual content in paragraphs 8a to 8f, they may not cover such images, audios and texts in view of the definitions of “film” and “sound recording” in Section 2 of the FA. Even if Sections 11(a) or 12(a) of the UPA, or Section 29 of the FA, were invoked, a question may arise as to who exactly “makes” the AIgenerated adult sexual content — the AI chatbot developer(s), deployer(s), the user, or a mix thereof? At this juncture, it is notable that it remains open whether the UPA and FA are fit for purpose to regulate such AI-generated sexual content — on one hand, it is arguable that such laws are tech-neutral or tech-agnostic, and hence applicable; on the other 56

Penal Code 1871 (Singapore) s 377BE(2). Criminal Law (Miscellaneous Amendments) Act 2025 (Singapore) s 49; Penal Code 1871 (Singapore) s 377BD(5); Ministry of Home Affairs, “Second Reading of the Criminal Law (Miscellaneous Amendments) Bill – Opening Speech by Ms Sim Ann, Senior Minister of State, Ministry of Foreign Affairs and Ministry of Home Affairs” (4 November 2025) <https://www.mha.gov.sg/media-room/newsroom/second-reading-of-the-criminal-lawmiscellaneous-amendments-bill-opening-speech/> (accessed 13 June 2026); Singapore Parliamentary Debates, Official Report (5 February 2025) vol 95 at col 22 (K Shanmugam, Minister for Home Affairs and Law). 58 Penal Code 1871 (Singapore) s 377BE(5). 59 Id, at s 377BD(8) . 60 Id, at ss 11(a), 11(b), 12(a) and 12(b). 57

139

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

hand, it is arguable that many of their definitions, especially in Section 2 of both Acts, are outdated largely because they were developed at a time when AI was not prevalent. 27. On civil platform regulation, Part 10A of the Broadcasting Act 1994 (“BA”), 61 its associated Codes of Practice for Online Safety such as those for Social Media Services (“SMS Code”) and App Distribution Services (“ADS Code”),62 and the Internet Code of Practice (“ICOP”),63 are relevant. Although the SMS Code may cover embedded AI chatbots on social media services (e.g. Grok on X, Meta AI on Facebook and Instagram), because Singapore’s Infocomm Media Development Authority (“IMDA”) has designated six social media services as “regulated online communication services” (“ROCs”) pursuant to Section 45K of the BA (i.e. Facebook, Instagram, X, YouTube, TikTok, HardwareZone);64 the ADS Code may cover standalone AI chatbots via apps, because Singapore’s IMDA has designated five app distribution services as ROCs pursuant to Section 45K of the BA (i.e. Apple App Store, Google Play Store, Huawei App Gallery, Microsoft Store and Samsung Galaxy Store);65 and the ICOP may cover standalone AI chatbots via web browsers, insofar as such AI chatbots are “programmes” that generated “prohibited material”, Singapore’s civil platform regulation legally permits AI chatbots to generate adult sexual content for adults’ “private” gratification for the following reasons. 28. First, the BA, SMS Code, ADS Code and ICOP may not cover embedded AI chatbots on messaging services and gaming services from both an ex ante66 and ex post67 perspective, and may not cover standalone AI chatbots via web browsers and apps from an ex ante68 perspective, especially as messaging services, gaming services, and standalone AI chatbots

As amended by the Online Safety (Miscellaneous Amendments) Act 2022 (Singapore). SMS Code, supra n 15; ADS Code, supra n 15. 63 Singapore, Internet Code of Practice <https://www.imda.gov.sg/-/media/imda/files/regulations-andlicensing/regulations/codes-of-practice/codes-of-practicemedia/policiesandcontentguidelinesinternetinternecodeofpractice.pdf> (accessed 13 June 2026). 64 Singapore, List of Designated Social Media Services <https://www.imda.gov.sg/-/media/imda/files/regulationsand-licensing/regulations/codes-of-practice/codes-of-practice-media/list-of-designated-social-media-services.pdf> (accessed 13 June 2026). 65 Singapore, List of Designated App Distribution Services <https://www.imda.gov.sg//media/imda/files/regulations-and-licensing/regulations/codes-of-practice/code-of-practice-app-distributionservices/list-of-designated-app-distribution-services.pdf> (accessed 13 June 2026). 66 Broadcasting Act 1994 (Singapore) ss 45M and 45N. 67 Id, at ss 45H, 45I and 45J. 68 Id, at ss 45M and 45N. 62

140

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

have not been designated as ROCs pursuant to Section 45K of the BA.69 Second, although the ADS Code imposed age assurance requirements on app stores effective from 31 March 2026, this age assurance requirement targets app stores instead of standalone AI chatbot developers. Moreover, age assurance requirements tend to lead to the conclusion that AI Chatbots are legally permitted to generate adult sexual content for adults’ “private” gratification — hence, the distinction drawn between child and adult users. Third, the fact that X’s rules and policies explicitly allow for “adult content”,70 and yet X continues to be permitted to operate for Singapore end-users, further strengthens my conclusion. 29. On remedies to victims, two statutory torts in Singapore’s recently passed Online Safety (Relief and Accountability) Act 2025 (“OSRAA”) may be particularly relevant — the torts of intimate image abuse and inauthentic material abuse.71 Moreover, the OSRAA amended, inter alia, sections 3 and 4 of the Protection from Harassment Act 2014 to recognise “humiliation” as a harm in addition to harassment, alarm and distress; and “sexual or indecent” words, behaviour and communication as a harm in addition to threatening, abusive and insulting words, behaviour and communication. 72 30. Overall, Singapore legally permits AI chatbots to generate the adult sexual content in paragraphs 8a to 8f for adults’ “private” gratification. D. United Kingdom 31. The UK appears to legally permit AI chatbots to generate adult sexual content for adults’ “private” gratification, albeit to a lesser degree than Australia, as explained below. 32. On criminal law, Section 138 of the Data (Use and Access) Act 2025, which amended the Sexual Offences Act 2003, likely covers the adult sexual content in paragraphs 8a and 8b, and may cover paragraphs 8c and 8d insofar as the AI chatbot-generated content constitutes a “purported intimate image of adult”, because it focuses on “creation”. 73 Furthermore, 69

Id, at ss 2, 2A and 45C. X Help Center, supra n 33. 71 Online Safety (Relief and Accountability) Act 2025 (Singapore) ss 13, 16, 83 and 86. 72 Online Safety (Relief and Accountability) Act 2025 (Singapore) s 109 and the Schedule; Singapore Parliamentary Debates, Official Report (5 November 2025) vol 96. 73 Sexual Offences Act 2003 (Singapore) ss 66E, 66F and 66G. 70

141

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

Section 138 goes one step further — it criminalises attempts to create adult sexual content, also known as “requesting the creation of purported intimate image of adult”, regardless of whether the purported intimate image was actually created. 74 However, it is notable that Section 138 unlikely covers the adult sexual content in paragraphs 8e and 8f, given that person B’s lack of consent is always relevant, 75 and likely only covers adult sexual content in paragraphs 8a to 8d in the formats of images and videos, and not audios and texts.76 Moreover, although Section 127 of the Communications Act 2003 likely covers the adult sexual content in paragraphs 8a to 8d, 77 it remains uncertain whether an AI chatbot provider can be said to have “caused an indecent or obscene message or matter to be sent” 78 such that the adult sexual content in paragraphs 8e and 8f are also covered. 79 Lastly, it is notable that Section 2 of the Obscene Publications Act 1959 unlikely covers the adult sexual content in paragraphs 8a to 8f because it requires “publication” of the obscene matter, unless the AI chatbot provider or the AI chatbot itself can constitute a “person who is likely to be depraved or corrupted”.80 33. On civil platform regulation, Ofcom’s guidance, 81 self-assessment tools, 82 and the wording of the UK Online Safety Act 2023 suggest that this Act does not generally cover standalone, embedded and user-created “private AI chatbots”, because such “private AI chatbots” often do not constitute “user-to-user services” under Part 3 of the Act,83 may not constitute “search services” under Part 3 of the Act, 84 and may not generate “pornographic content” under Part 5 of the Act. 85 To quote Ofcom, “AI-generated content is only covered by the 74

Id, at s 66F. Id, at ss 66E, 66F and 66G. 76 Id, at ss 66A, 66D, 66E, 66F and 66G. 77 Clare McGlynn et al, supra n 13, at p 76. 78 Communications Act 2003 (UK) s 127(1)(b). 79 Clare McGlynn et al, supra n 13. 80 Obscene Publications Act 1959 (UK) s 1; Clare McGlynn et al, supra n 13, at p 75. 81 Ofcom’s open letter, supra n 12; Ofcom, “Ofcom update: Investigation into X, and scope of the Online Safety Act” (3 February 2026) <https://www.ofcom.org.uk/online-safety/illegal-and-harmful-content/investigation-into-xand-scope-of-the-online-safety-act> (accessed 13 June 2026) (“Ofcom update”); Ofcom, “AI chatbots and online regulation – what you need to know” (18 December 2025) <https://www.ofcom.org.uk/online-safety/illegal-andharmful-content/ai-chatbots-and-online-regulation-what-you-need-to-know> (accessed 13 June 2026) (“Ofcom on AI chatbots”). 82 Ofcom, “Check if the Online Safety Act applies to you” (27 February 2024) <https://www.ofcom.org.uk/onlinesafety/illegal-and-harmful-content/check> (accessed 13 June 2026) (“Ofcom checker”). 83 Online Safety Act 2023 (UK) ss 3 and 4. 84 Ibid. 85 Ofcom update, supra n 81; Ofcom checker, supra n 82; Clare McGlynn et al, supra n 13, at p 16. 75

142

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

illegal content and children’s safety duties in Part 3 of the Act if it is ‘user-generated’ (shared by users with each other) or ‘search content’ (encountered in or via search results). Generation of other chatbot content – such as in a one-to-one interaction between a user and a chatbot that does not involve searching the internet or sharing with other users – is not regulated under Part 3, but could be under Part 5 if it is pornographic”. 86 34. Notably, there are complications surrounding the question of whether “private AI chatbots” constitute “user-to-user services” and “search services” under Part 3 of the Act. a. On “user-to-user services”, as per a research project funded by the UK Research and Innovation, “services, such as Character.AI, that allow users to create ‘persona’ chatbots and share them with other users would constitute user-to-user services, according to Ofcom: Where a site or app allows users to upload or create their own Generative AI chatbots – ‘user chatbots’ – which are also made available to other users, it is also a user-to-user service. This includes services that provide tools for users to create chatbots that mimic the personas of real and fictional people, which can be submitted to a chatbot library for others to interact with. Any text, images or videos created by these ‘user chatbots’ is ‘user-generated content’ and is regulated by the Act”.87 b. On “search services”, as per a research project funded by the UK Research and Innovation, “chatbots that embed results from a search index in response to user queries would fall within the scope of a ‘search service’, but only in relation to content that comes from the search index. So, for example, ChatGPT would fall within the scope of the OSA in relation to some content it produces, but not all”. Furthermore, “[a] chatbot that provides search results from a search engine (e.g. by drawing on a third party’s search index) would be considered a search service. However, only the results from that search engine would amount to regulated search content. Any content coming from the model part of the chatbot would not be search content… [T]his is where it can get complicated; an outside observer hearing that

86 87

Ofcom update, supra n 81. Clare McGlynn et al, supra n 13, at pp 76-77.

143

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

a chatbot is in scope of the Act as a search service would understandably expect everything being produced by that chatbot to be regulated. However, this isn’t necessarily the case”. 88 35. If “private AI chatbots” constitute neither “user-to-user services”89 nor “search services”90 under Part 3 of the Act, they will unlikely be regulated under the Act unless they generate “pornographic content”91 under Part 5 of the Act, as also noted in Ofcom’s Guidance on Content Harmful to Children.92 However, even if a “private AI chatbot” is regulated under Part 5 of the Act, its key obligation will be to implement highly effective Age Assurance, 93 which highlights how the UK Online Safety Act 2023 does not legally prohibit AI Chatbots from generating adult sexual content for adults’ “private” gratification. Furthermore, the fact that X’s rules and policies explicitly allow for “adult content”, 94 and yet X continues to be permitted to operate for end-users in the UK, strengthens my conclusion that the UK legally permits AI chatbots to generate adult sexual content for adults’ “private” gratification. 36. Moreover, in light of the recent Grok fiasco, 95 the UK government conceded that certain AI chatbots may fall beyond the scope of the UK Online Safety Act 2023 as evident from the following three quotes.96 First, that “[t]he [UK] government will move fast to shut a legal loophole and force all AI chatbot providers to abide by illegal content duties in the Online Safety Act or face the consequences of breaking the law”. 97 Second, that the UK “[g]overnment [was] tabling an amendment to the Crime and Policing Bill to allow the government to require chatbots not currently in scope of the Online Safety Act to protect 88

Clare McGlynn et al, supra n 13, at p 15. Online Safety Act 2023 (UK) ss 3 and 4. 90 Ibid. 91 Ofcom checker, supra n 82; Clare McGlynn et al, supra n 13 at p 16. 92 Ofcom, Protecting Children from Harms Online: Guidance on Content Harmful to Children (24 April 2025) <https://www.ofcom.org.uk/siteassets/resources/documents/consultations/category-1-10-weeks/statementprotecting-children-from-harms-online/main-document/guidance-on-content-harmful-to-children.pdf?v=395445> (accessed 13 June 2026). 93 Ofcom on AI chatbots, supra n 81. 94 X Help Center, supra n 33. 95 As briefly explained in paragraph 3 above. 96 United Kingdom Government, “PM: “No platform gets a free pass”: Government takes action to keep children safe online” (15 February 2026) <https://www.gov.uk/government/news/pm-no-platform-gets-a-free-passgovernment-takes-action-to-keep-children-safe-online> (accessed 13 June 2026). 97 Ibid. 89

144

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

their users from illegal content”. 98 Third, in response to the Grok fiasco on X, Ofcom noted that “[b]ecause of the way the Act relates to chatbots… we are currently unable to investigate the creation of illegal images by the standalone Grok service in this case”. 99 Overall, the UK’s civil platform regulation likely legally permits AI chatbots to generate adult sexual content, in paragraphs 8a to 8f, for adults’ “private” gratification. 37. On remedies for victims, although the UK government has invested much effort to advance its Violence Against Women and Girls national strategy, 100 the UK Criminal Injuries Compensation Scheme may not enable victims of non-consensual intimate image abuse to obtain compensation.101 As per the Women and Equalities Committee in a House of Commons committee report, “[t]he list of offences that are within scope of the Criminal Injuries Compensation Scheme is out of date. Crimes perpetrated online, such as nonconsensual intimate image abuse, can be just as damaging to a person as those involving physical violence. They can have a catastrophic impact on a person’s mental health. It is essential that victims of such crimes are able to access compensation”. 102 Moreover, as per a research project funded by the UK Research and Innovation, “proving the requisite intent

Ibid; United Kingdom, House of Commons, Oral evidence: Work of the Secretary of State for the Department for Science, Innovation and Technology, HC 1543 (3 December 2025) (Liz Kendall MP, Secretary of State, Department for Science, Innovation and Technology); United Kingdom, Parliament, Artificial Intelligence: Safety Question for Home Office UIN HL15942 (13 April 2026) (Lord Hanson of Flint); United Kingdom Government, “Police use of artificial intelligence (AI): factsheet (accessible)” (9 June 2026) <https://www.gov.uk/government/publications/police-use-of-artificial-intelligence-ai-factsheet/police-use-ofartificial-intelligence-ai-factsheet-accessible> (accessed 28 June 2026); Heather Catchpole & Oliver Carroll, “UK Government Children's Safety and AI Chatbot Powers: Two New Acts Receive Royal Assent” (30 April 2026) <https://www.twobirds.com/en/insights/2026/uk/uk-government-children's-safety-and-ai-chatbot-powers-two-newacts-receive-royal-assent> (accessed 28 June 2026). 99 Ofcom update, supra n 81. 100 Crown Prosecution Service, “Violence Against Women and Girls Strategy 2025-2030” <https://www.cps.gov.uk/publication/vawg-2025-2030> (accessed 13 June 2026); Helen Frowe & Jonathan Parry, “How to prevent consumption of non-consensual intimate images” (13 March 2026) <https://blogs.lse.ac.uk/politicsandpolicy/how-to-prevent-consumption-of-non-consensual-intimate-images/> (accessed 13 June 2026) (“Frowe & Parry on NCII”); Helen Frowe & Jonathan Parry, “The case for criminalising revenge porn consumption” (15 March 2022) <https://blogs.lse.ac.uk/politicsandpolicy/criminalising-revenge-pornconsumption/> (accessed 13 June 2026); Jonas Haeg, “Grok Doesn’t Degrade Women—You Do! Deepfakes, Responsibility and Language” (18 January 2026) <https://www.publicethics.org/post/deepfakes-responsibility-andlanguage> (accessed 13 June 2026). 101 United Kingdom, House of Commons committee report on Tackling non-consensual intimate image abuse (2025). 102 Ibid.

145

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

on the part of the chatbot provider” under the Protection from Harassment Act 1997, and proving “harassment” under the Equality Act 2010, will be challenging. 103 38. Overall, the UK appears to legally permit AI chatbots to generate adult sexual content in paragraphs 8a to 8f for adults’ “private” gratification, albeit to a lesser degree than Australia. III. APPLICATION OF JURISPRUDENTIAL THEORIES TO THE QUESTION OF WHETHER AI CHATBOTS SHOULD BE LEGALLY PERMITTED TO GENERATE SEXUAL CONTENT FOR ADULTS’ “PRIVATE” GRATIFICATION 39. I will now strive to apply jurisprudential theories to the question of whether AI chatbots should be legally permitted to generate sexual content for adults’ “private” gratification. Specifically, I will analyse the following four buckets of jurisprudential theories: (a) John Stuart Mill, (b) Hart-Devlin debate, (c) Rawls-Finnis debate, and (d) Virtue Ethics. While bucket (b) focuses on criminal law, the other three buckets cover all regulations generally. A. John Stuart Mill 40. John Stuart Mill believes that the coercive force of the law by a state against an individual is only justifiable to “prevent harm to others”, and the state should not regulate “selfregarding” acts which only physically or morally harm the actor alone and no one else, to facilitate everyone’s pursuit of liberty. 104 Mill elucidated this through his “harm principle” which stated, inter alia, that “the sole end for which mankind are warranted, individually or collectively, in interfering with the liberty of action of any of their number, is selfprotection. That the only purpose for which power can be rightfully exercised over any member of a civilised community, against his will, is to prevent harm to others. His own good, either physical or moral, is not a sufficient warrant… The only part of the conduct of anyone for which he is amenable to society is that which concerns others. In the part which merely concerns himself, his independence, is, of right, absolute. Over himself, over

103 104

Clare McGlynn et al, supra n 13, at pp 53-54. John Stuart Mill, On Liberty (Batoche Books, 2001) at p 13.

146

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

his own body and mind, the individual is sovereign”. 105 Unlike Devlin,106 Mill effectively suggests that “societal morality” is not a legitimate justification for the coercive force of the law to be used against an individual. 41. Applying Mill’s harm principle to the present question, Mill would likely suggest that paragraphs 8a to 8f should be legally permissible, given that there is no further publication or dissemination of the AI chatbot-generated adult sexual content beyond the one user, and the law should only interfere with one’s liberty to “prevent harm to others”. At its strictest, although unlikely, Mill’s harm principle may suggest that paragraphs 8a to 8d are legally impermissible given that the “non-consensual” element morally harms the victim, 107 but would highly unlikely suggest that paragraphs 8e to 8f should be legally impermissible. 42. At this juncture, I note that Mill’s theory suffers from numerous problems. 43. First, Mill’s definition of “harm” is ambiguous — while some argue that Mill’s harm principle is only confined to direct physical harm, such that regulation is only justified to prevent direct physical harm to third parties, others argue that Mill’s harm principle applies to both physical and moral harm, such that regulation is justified to prevent physical or moral harm to third parties.108 Moreover, it remains highly debatable whether one can draw clear-cut distinctions between “physical harm” and “moral harm”, “direct physical harm” and “indirect physical harm”, the degree of causality required to establish the applicability of the harm principle, etc. This is so especially because, inter alia, research has shown that consuming pornographic content places one at an increased risk of developing sexually deviant tendencies and committing sexual offences,109 which disproportionately harms 105

Ibid. As expounded on in paragraphs 47 to 54 below. 107 Ben Saunders, “Reformulating Mill’s Harm Principle” (2016) Mind 1005-1032; John Stuart Mill, The Subjection of Women (1869); Clare McGlynn & Ian Ward, “Would John Stuart Mill have Regulated Pornography?” (2014) Journal of Law and Society 41(4) 500-522; Robert Skipper, “Mill and Pornography” (1993) Ethics 103(4) 726-730. 108 David Dyzenhaus, “John Stuart Mill and the Harm of Pornography" (1992) 102 Ethics 534 at 544-546; Robert Skipper, supra n 107; Michael Joel Kessler, “A Puzzle about Obscenity” (2017) De Ethica 4(1) 5-30; Henry Fradella & Jennifer Sumner, Sex, Sexuality, Law, and (In)justice (Routledge, 2016) at p 88. 109 Matt Fradd, The Porn Myth: Exposing the Reality Behind the Fantasy of Pornography (Ignatius Press, 2017); Jill C. Manning, “The Impact of Internet Pornography on Marriage and the Family: A Review of the Research” (2006) Sexual Addiction and Compulsivity 13(2) 131-165; James V. P. Check & Ted H. Guloien, Reported Proclivity for Coercive Sex Following Repeated Exposure to Sexually Violent Pornography, Nonviolent Dehumanizing Pornography, and Erotica (Routledge, 1989); Mimi H. Silbert & Ayala M. Pines, “Pornography and sexual abuse of women” (1984) Sex Roles 857-868; Andrew R. Chow & Solcyré Burga, “Chatbots Are Becoming More Sexually 106

147

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

women and children.110 Moreover, as per Robert George, “[t]he central harm of pornography is not… that it shocks and offends people, any more than the central harm of carcinogenic smoke is that it smells bad. Rather, the central harm of pornography is moral harm — harm to character, and thus to the human goods and institutions, such as the good and institution of marriage, which are preserved and advanced by the disposition to act uprightly, and damaged and defiled by a contrary disposition, in respect to them”. 111 44. Second, Mill’s scope of “self-regarding” and “private” acts is ambiguous — following Tan Seow Hon, should “the category of prohibited acts [encompass] acts that affect others, or only acts that affect the interests of others”? 112 If it is the former, it would seem that all acts do “affect others” insofar as our acts affect us and we inevitably affect others. 113 This view could also suggest that there are no “purely self-regarding” acts as per Mill, and no realm of “private immorality” as per Hart and Devlin. 114 If it is the latter, a question arises as to what types of “interests” are relevant. 115 Moreover, as alluded to in paragraph 43 above, even seemingly “self-regarding” and “private” acts may inevitably and subconsciously harm others especially in the context of marriage, families and human relationships. For example, the habitual consumption of sexual content for one’s “private” gratification may destroy marriages and families by rewiring one’s dopamine reward system in a manner

Explicit in a Bid to Attract Usership and Paying Customers” (16 October 2025) <https://time.com/7326111/chatgptopenai-explicit-erotica-update/> (accessed 13 June 2026). 110 United Kingdom Government, “The relationship between pornography use and harmful sexual attitudes and behaviours: literature review” (15 January 2021) <https://www.gov.uk/government/publications/the-relationshipbetween-pornography-use-and-harmful-sexual-behaviours/the-relationship-between-pornography-use-and-harmfulsexual-attitudes-and-behaviours-literature-review> (accessed 13 June 2026); Ysa Marie, supra n 5; Furizal et al, “Social, legal, and ethical implications of AI-Generated deepfake pornography on digital platforms: A systematic literature review” (2025) Social Sciences & Humanities Open; Office of the United Nations High Commissioner for Human Rights, “Internet Pornography Harms Women and Girls” <https://www.ohchr.org/sites/default/files/documents/issues/women/sr/cfis/new-frontier/subm-sr-violence-againstcso-71-concept-truth.pdf> (accessed 13 June 2026); Meghan Donevan, “Exposing pornography's true nature: A conceptual framework” (2025) Women's Studies International Forum; Daniel Thomas, “Ad for AI editing app which said it could 'remove anything' banned” (18 March 2026) <https://www.bbc.com/news/articles/cx2g8888q53o?at_medium=RSS&at_campaign=rss> (accessed 13 June 2026). 111 Robert George, “The Concept of Public Morality” (2000) 45 American Journal of Jurisprudence 17 at 18. 112 Tan Seow Hon, “Pragmatism, Morals Legislation and Criminalization of Homosexual Acts in Singapore” (2009) Journal of Comparative Law 3(2) 285-304 at 287. 113 Ibid. 114 John Stuart Mill, supra n 104, at pp 75 and 89; H. L. A. Hart, Law, Liberty, and Morality (Stanford University Press, 1963) at pp 38-48 and 52; Patrick Devlin, The Enforcement of Morals (Oxford University Press, 1965) at pp 3, 15-16, 19 and 109-112. 115 Tan Seow Hon, supra n 112, at 287.

148

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

akin to substance addiction, thereby creating unrealistic or warped expectations of “sexual satisfaction” that cannot be practically fulfilled within healthy marriages. Another example, a husband’s habitual consumption of sexual content for his “private” gratification may lead him to use his wife as a sexual object, devoid of human dignity, solely for his pleasure, instead of loving her as an end in herself in sexual acts. In essence, Mill’s distinction between “self-regarding” and “other-regarding” acts appears rather arbitrary. 45. Third, more fundamentally, Mill fails to justify why liberalism, above and beyond all other values, should be the key value that is prioritised in all legal systems, and arguably harbours a distorted understanding of “liberty”. This is evident from, inter alia, two facts. First, Mill draws a false dichotomy between “liberty” and the “prevention of harm” to one’s self or others — he falsely suggests that the prevention of moral harm to one’s self and others does not enhance liberty, and in fact, impairs liberty. Mill’s understanding of liberty, autonomy and freedom evidently lie in stark contrast to that of virtue ethicists, as expounded on in bucket (d) below. Second, as also noted by Hart, Mill’s rejection of paternalism overlooks the fact that not everyone has “the psychology of a middle-aged man whose desires are relatively fixed”, because “[c]hoices may be made or consent given without adequate reflection or appreciation of the consequences; or in pursuit of merely transitory desires; or in various predicaments when the judgment is likely to be clouded”. 116 In essence, paternalism may be justified especially when people lack true free will. Moreover, Mill’s harm principle has far-reaching consequences — for instance, it may fail to justify laws requiring motorcyclists to wear helmets, 117 laws requiring passengers to wear seatbelts, etc. Overall, although Mill expressed concerns about the “tyranny of the majority”, it is arguable that his utilitarian theory ultimately promotes such “tyranny” albeit under the guise of “individual liberty”. 118

116

H. L. A. Hart, supra n 114, at p 136. Robert George, In Defense of Natural Law (Oxford University Press, 1999) at p 312. 118 John Stuart Mill, Utilitarianism (Bibliotech Press, 2020). 117

149

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

B. Hart-Devlin debate 46. The Hart-Devlin debate emerged in the 20th century, concerning the key question of the extent to which morality should be enforced through criminal law. I will first analyse Devlin’s theory, before analysing Hart’s theory. 47. As per Patrick Devlin, criminal law has always been concerned with the enforcement of morality, a “public morality” exists because society is a “community of ideas”, and the “suppression of vice is as much the law's business as the suppression of subversive activities”.119 Specifically, an act should be criminalised if it violates the “moral code”, “moral judgements” or “common morality” of society, as ascertained by a “reasonable man”, and the “limits of tolerance” must have been reached, even if the act causes no physical harm to individuals and society, because such criminalisation will prevent society from “disintegration”.120 Moreover, Devlin endorsed three key working principles. 121 First, that “the law is concerned with the minimum and not with the maximum”. In essence, Devlin did not equate sin with crime, and believed that the maximum individual freedom that is consistent with the integrity of society should be allowed at least in matters of conscience — criminal law deals with the minimum standards of human conduct while moral law deals with the maximum standard of human conduct. Second, that “in any new matter of morals the law should be slow to act”. As the limits of tolerance in a society tends to shift over time, one should discern whether each shift is a lasting or merely temporal one. Third, that “as far as possible privacy should be respected”. In essence, “the right to privacy is something to be put in the balance against the enforcement of the law”, such that morals should only be enforced when they are necessary to protect the “moral code” of society. 48. Applying Devlin’s theory to the present question, Devlin may suggest that paragraphs 8a to 8d should be criminalised, especially because the “non-consensual” element arguably suggests that the “limits of tolerance” have been reached as there a “real feeling of

119

Patrick Devlin, The Enforcement of Morals (Oxford University Press, 1965) at p 14. Id, at pp 10, 11, 13, 15, 17-18, 32 and 115. 121 Id, at p 19. 120

150

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

reprobation” and “deeply felt unmanufactured disgust” are present. 122 Alternatively, Devlin may suggest that paragraphs 8a to 8f should not be criminalised, especially in view of his three working principles and societal “disintegration” thesis. Regardless, it is highly unlikely that Devlin’s theory would suggest that paragraphs 8e to 8f should be criminalised. 49. At this juncture, I note that Devlin’s theory suffers from numerous problems. 50. First, who is the supposed “reasonable man” or “right-minded man”, who is not the “rational” or “good” man, but the “man in the Clapham omnibus”? 123 Are his views based solely on subjective feelings and sentiments, as Devlin’s quote that “his judgement may be largely a matter of feeling” seems to suggest, 124 or does he believe in the existence of an objective moral truth? The difficulty in identifying the “reasonable man” will contribute to difficulties in ascertaining the “moral judgements of society” and where the “limits of tolerance” lie. 51. Second, are the concepts of “unmanufactured disgust” and “reprobation” based solely on subjective feelings and sentiments, or must they be rooted in objective morality? Similarly, the difficulty in defining these vague and amorphous concepts accurately will contribute to difficulties in ascertaining whether the “limits of tolerance” have been reached. 52. Third, Devlin’s “disintegration” thesis may appear far-fetched — will a change in moral values truly result in the disintegration of society or breakdown of social order? 125 Nevertheless, following Robert George, it is arguable that Devlin’s “disintegration” thesis was more concerned about the protection of societal moral ecology, or the fundamental principles on which laws pivot, from moral slippery slopes to uphold the community’s common good for human flourishing. 126 53. Fourth, although Devlin opined that criminal law should not criminalise on the basis of majority disapproval, it is arguable that his theory, which relies on the ambiguous concept 122

Id, at p 17. Id, at p 15. 124 Id, at p 15. 125 Tan Seow Hon, “The Case Against Physician-Assisted Suicide and Voluntary Active Euthanasia” (2017) 29 SAcLJ 375. 126 Ibid; Robert George, Making Men Moral: Civil Liberties and Public Morality (Clarendon Press, 1995) at p 66. 123

151

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

of the “reasonable man”, and the requirement that “the moral judgement of society must be something about which any twelve men or women drawn at random might after discussion be expected to be unanimous”, 127 suggests that Devlin’s utilitarian theory ultimately criminalises on the basis of majority disapproval albeit under the guise of the “moral code”, “moral judgements” or “common morality” of society. 54. Fifth, it remains uncertain whether Devlin is a moral non-cognitivist, as evident from his emphasis on mere sentiments and feelings in his definition of a “reasonable man” and “unmanufactured disgust”, or a moral objectivist, because he appears to believe that there is a minimal morality that binds everyone as evident from his quote on a society of drunkards and gamblers reflected below. a. As per Devlin, “I think there can be no theoretical limits to legislation against immorality. You may argue that if a man's sins affect only himself it cannot be the concern of society. If he chooses to get drunk every night in the privacy of his own home, is anyone except himself the worse for it? But suppose a quarter or a half of the population got drunk every night, what sort of society would it be? You cannot set a theoretical limit to the number of people who can get drunk before society is entitled to legislate against drunkenness. The same may be said of gambling. The Royal Commission on Betting, Lotteries, and Gaming took as their test the character of the citizen as a member of society. They said: 'Our concern with the ethical significance of gambling is confined to the effect which it may have on the character of the gambler as a member of society. If we were convinced that whatever the degree of gambling this effect must be harmful we should be inclined to think that it was the duty of the state to restrict gambling to the greatest extent practicable.'”.128 b. At this juncture, it is notable that Australia’s eSafety has drawn parallels between “age restrictions for pornography” and “age restrictions for… alcohol and public gambling”,129 and the UK’s Ofcom has drawn parallels between laws in society 127

Patrick Devlin, supra n 119, at p 15. Id, at p 14. 129 eSafety FAQs, supra n 15. 128

152

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

which have “long protected youngsters from things like alcohol… and gambling” and laws in the digital world to protect youngsters from “online pornography”. 130 Although these quotes relate to child users, unlike my present question which is targeted at adult users, the parallels drawn between online sexual content, alcohol and gambling arguably suggest that Devlin could potentially extend or analogise his quotation in paragraph 54a to my present situation of AI chatbots generating adult sexual content for adults’ “private” gratification. This is so because, following Robert George and as expounded in paragraphs 43 and 44 above, the central harm of pornography is its “harm to character”, and its denigration of the goods and institutions of marriage and family in society, especially if consumed habitually as such consumption rewires one’s dopamine reward system in a manner akin to drug, alcohol and gambling addiction.131 Moreover, many AI chatbot developers are increasingly introducing “paid subscriptions”, where premium subscribers can pay to unlock more paywalled features to create and consume more highly sexualised AI chatbot-generated content, 132 thereby exacerbating all harms associated with such “private” gratification. c. Overall, as per Tan Seow Hon, one may criticise Devlin’s “willingness to protect any society from his perceived threat of disintegration without regard to whether the moral code of the society in question is salutary or bad”. 133 55. H.L.A. Hart essentially strived to minimise the extent to which morality should be enforced through criminal law by criticising Devlin’s theory and modifying Mill’s theory. As per Hart, criminal law is not deeply connected with morality — the fact of “positive morality”

130

Ofcom, “4chan fined £450,000 for not protecting children from online pornography” (19 March 2026) <https://www.ofcom.org.uk/online-safety/illegal-and-harmful-content/4chan-fined-450000-for-not-protectingchildren-from-online-pornography> (accessed 13 June 2026). 131 Robert George, supra n 111; Matt Fradd, supra n 109. 132 Matt Burgess, “X Didn’t Fix Grok's ‘Undressing’ Problem. It Just Makes People Pay for It” (9 January 2026) <https://www.wired.com/story/x-didnt-fix-groks-undressing-problem-it-just-makes-people-pay-for-it/> (accessed 28 June 2026); Hadas Gold, “After ‘digital undressing’ criticism, Elon Musk’s Grok limits some image generation to paid subscribers” (10 January 2026) <https://edition.cnn.com/2026/01/09/business/grok-image-generationundressing-deepfake> (accessed 28 June 2026); Ashley Belanger, “Musk’s tactic of blaming users for Grok sex images may be foiled by EU law” (19 March 2026) <https://arstechnica.com/tech-policy/2026/03/eu-moves-to-bannudify-apps-after-grok-made-them-mainstream/> (accessed 13 June 2026); Ysa Marie, supra n 5. 133 Tan Seow Hon, supra n 125; Robert George, supra n 126, at p 71.

153

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

of society as such may be insufficient to justify its enforcement in law, criminal law should not be used purely to enforce the “positive morality” of society, and Devlin’s moderate thesis, which states that “a shared morality is the cement of society; without it there would be aggregates of individuals but no society”, is “absurd” for suggesting that “a change in [society’s shared] morality is tantamount to the destruction of a society”. 134 Moreover, Hart added two qualifications to Mill’s harm principle. 135 First, paternalism as a justification for criminal law in relation to physical harm, as also noted in paragraph 45 above, as numerous individuals do not know our own interests best. Second, public offence as a justification for criminal law, such that laws should regulate acts of “public indecency” but not acts of “private immorality”. To quote Hart, “a right to be protected from the distress which is inseparable from the bare knowledge that others are acting in ways you think wrong, cannot be acknowledged by anyone who recognises individual liberty as a value”. 136 56. Applying Hart’s theory to the present question, Hart would likely suggest that paragraphs 8a to 8f should not be criminalised, especially because law is not concerned with the enforcement of morals, and his concept of paternalism appears to be confined to physical harm. Nevertheless, even if the application of Hart’s theory lends itself to the same result as the application of Devlin’s theory to the present question, it is notable that their bases for reaching the same conclusion are starkly different. 57. At this juncture, I note that Hart’s theory suffers from numerous problems. 58. First, Hart’s paternalism may ultimately rest on morals. As per Tan Seow Hon, “[t]his is because law only protects, and consensus only requires law to protect, individuals from morally reprehensible conduct that leads to physical harm”. 137 For example, would the criminalisation of online gambling be justifiable on the basis of its propensity to cloud one’s judgement in the pursuit of merely transitory desires vis-a-vis physical harm, or on the basis of the immorality of online gambling? Moreover, the determination of how “clouded” one’s judgment is, or whether one is “in pursuit of merely transitory desires”,

134

H. L. A. Hart, supra n 114, at pp 20, 22, 33 and 48-52. Id, at pp 30-34 and 38-48. 136 Id, at p 46. 137 Tan Seow Hon, supra n 112, at 288. 135

154

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

may ultimately hinge on morality. Thus, it is arguable that one cannot decide on the issue of criminalisation without first having engaged morality. 138 59. Second, Hart’s distinction between “publicly offensive” acts and “privately immoral” acts may be arbitrary, especially as the determination of the degree of “public offensiveness” or “public indecency” may ultimately hinge on the majority’s subjective prejudices and biases. As per Tan Seow Hon, “[u]nless one is prepared to refer either to the moral opinions of the majority or morality in some form or other, one cannot adequately explain why one should not take into account the offence to feminists of viewing posters of scantily-clad women, or to some teetotalers seeing people drink in public, and so on”. 139 Overall, Hart’s utilitarian theory ultimately criminalises on the basis of majority disapproval albeit under the guise of “public offensiveness” and “public indecency”. C. Rawls-Finnis debate 60. The Rawls-Finnis debate emerged in the 20th century, concerning the key question of what types of reasons are legitimate to justify laws. I will first analyse Rawls’s theory, before analysing Finnis’s theory. 61. Under John Rawls’s theory of political liberalism, Rawls prioritised the “right” over the “good” by suggesting that in a society marked by “reasonable pluralism”, with different and conflicting “comprehensive doctrines” where there are disagreements about the “good”, no single comprehensive doctrine is appropriate as the freestanding “political conception of justice”. 140 As per Rawls, the “political conception of justice” which comprises “political values” that all citizens “may reasonably be expected to endorse”, is stabilised by an “overlapping consensus” of “reasonable” comprehensive doctrines, and evinces how “public reason” serves as a justification for legitimate law. 141 In essence, Rawls opined that political institutions should grant each individual maximum autonomy and liberty to pursue their own conception of the good life — by purporting to refuse to settle the metaphysical question of the “good”, and promoting “reasonableness”, 138

Id, at 288. Id, at 288-289. 140 John Rawls, Political Liberalism (Columbia University Press, 1996) Lectures I, IV, V, VI and IX. 141 Ibid. 139

155

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

“reciprocity” and a “readiness to meet others halfway”, Rawls claimed to be “neutral” amongst different conceptions of the good. 62. Applying Rawls’s theory to the present question, Rawls would likely suggest that paragraphs 8a to 8f should be legally permissible to maximise individual liberty, avoid settling the metaphysical question of the “good”, and be “neutral” amongst different conceptions of the good. At its strictest, although unlikely, Rawls’s theory may suggest that paragraphs 8a to 8d should be legally impermissible in view of his concepts of “original position” and the “veil of ignorance”, 142 but would highly unlikely suggest that paragraphs 8e to 8f should be legally impermissible. 63. At this juncture, I note that Rawls’s theory suffers from numerous problems. 64. First, although Rawls claims to be “neutral” amongst different conceptions of the good by, inter alia, purporting to leave the metaphysical question of the “good” unsettled, this is untrue for two reasons. a. First, critics like Neal have argued that Rawls’s claim to be “neutral” amongst different conceptions of the good is in and of itself a “distinct conceptualization of what it means to have a conception of the good”.143 Thus, Rawls’s theory of political liberalism and reasonable pluralism may, in and of itself, be a “comprehensive doctrine” that fails to be neutral on the conception of the good — Rawls’s conception of the good is to accord each individual the right to define his/her own conception of the good, or to give primacy to self-determination of the good. This criticism is further strengthened by Rawls’s labelling and rejection of “unreasonable” comprehensive doctrines, labelling and sidelining of “nonpolitical”

142

John Rawls, A Theory of Justice (Belknap Press, 1971); Ibid; Kyle William Chapel, “Rawlsian Self-Respect and Limiting Liberties in the Background Culture” (May 2016) <https://trace.tennessee.edu/server/api/core/bitstreams/eb49110c-1bbf-4a61-80e5-f15419e12cc0/content> (accessed 13 June 2026); Caroline West, “Pornography and Censorship” (5 May 2004) <https://plato.stanford.edu/archives/fall2024/entries/pornography-censorship/> (accessed 13 June 2026); Kimberly A. Yuracko, “Toward Feminist Perfectionism: A Radical Critique of Rawlsian Liberalism” (1995) UCLA Women's Law Journal 6(1); Morten Bay, “AI Ethics and Policymaking: Rawlsian Approaches to Democratic Participation, Transparency, Accountability, and Prediction” (2023). 143 Patrick Neal, A Liberal Theory of the Good? (1987) Canadian Journal of Philosophy 17(3) 567-581 at 571-572.

156

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

values, and attempts to sidestep complex and contentious questions relating to the good or morals. b. Second, Rawls often settles the metaphysical question of the “good” by, inter alia, sidelining metaphysical arguments which draw upon religion and morality by branding them as “unreasonable” comprehensive doctrines, such that Rawls’s theory almost always tends towards permissive laws.144 Specifically in this context, Rawls would have settled the metaphysical question of the moral worth or value of human dignity in the human body by arguing that AI chatbots should be legally permitted to generate adult sexual content for adults’ “private” gratification, and that those who strongly object to such practices need not themselves exercise their right to generate such content for “private” gratification, but should nonetheless recognise these laws as legitimate. 145 65. Second, Rawls’s delineation between the “political domain” and “nonpolitical domain” may be superficial, especially because both domains may occasionally overlap. For example, Rawls’s dismissive treatment of “rationalist believers” as “mistaken in denying that fact [of reasonable pluralism]”, because “religious beliefs [cannot] be publicly and fully established by reason” and without further investigation into the merits of the truth claims by such rationalist believers,146 evinces how Rawls’s reliance on political values in his theory renders any reference to religious perspectives nugatory despite overlaps between political and non-political values. 66. Third, it is uncertain how Rawls evaluates or determines what “all citizens as free and equal may reasonably be expected to endorse”, what constitutes “unreasonable” comprehensive doctrines and an “overlapping consensus”, and what it means to demonstrate “reasonableness”, "reciprocity" and to “meet others halfway”. For example, although Rawls noted that “unreasonable” comprehensive doctrines are comprehensive doctrines which “do not gain enough currency to undermine society’s essential justice” and may “die

144

John Rawls, supra n 140, at Lectures II and VI. Tan Seow Hon, supra n 125. 146 John Rawls, supra n 140, at pp 152 and 153; John Finnis, “Public Reason, Abortion, and Cloning” (1998) Valparaiso University Law Review 361-382 at 366. 145

157

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

out”,147 the standard or criteria that Rawls ultimately relies upon to assess the degree of reasonableness of each comprehensive doctrine remains ambiguous. Moreover, Rawls’s concept of “overlapping consensus” is ultimately superficial because it provides little guidance as to how one can identify the shared political values amongst citizens in a particular society. Furthermore, although Rawls opined that “[t]he criterion of reciprocity requires that when [terms of cooperation] are proposed as the most reasonable terms of fair cooperation, those proposing them must also think it at least reasonable for others to accept them, as free and equal citizens”, 148 one may argue that this requirement for the terms of cooperation to be “most reasonable” to the proposer but “at least reasonable” to others may inevitably lead to a situation where the majority’s comprehensive doctrine is always favoured over the minority’s to justify laws, such that Rawls’s theory is ultimately utilitarian. 67. As per John Finnis, laws can be justified if they promote human flourishing, and the coercive force of the state is justified if it aims to restrict an individual from a rationally unintelligible or morally impermissible act. 149 While the pursuit of the seven self-evident basic human goods makes one’s actions rationally intelligible, fulfilling the nine principles of practical reasonableness makes one’s actions morally permissible. 150 Furthermore, Finnis noted that one should never choose an act which “itself damages and itself does not promote some basic good”, and “every basic [good must] be at least respected in each and every action”, although Finnis recognised that the “concentration of effort will indirectly impoverish, inhibit, or interfere with the realization of… other [basic goods]”.151 In essence, the coercive force of the state should facilitate our pursuit of the seven basic human goods, nine principles of practical reasonableness, and human flourishing. Unlike Rawls, Finnis only allows for autonomy within reason and rejects the type of autonomy defined by Kennedy J in Planned Parenthood v Casey — “[a]t the heart of liberty is the

147

John Rawls, supra n 140, at pp 39 and 197. John Rawls, “The Idea of Public Reason Revisited” (1997) The University of Chicago Law Review 64(3) 765807 at 770. 149 John Finnis, Natural Law & Natural Rights (Oxford University Press, 2nd Ed, 2011) Part Two. 150 Id, at Part Two. 151 Id, at pp 119-120. 148

158

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

right to define one’s own concept of existence, of meaning, of the universe, and of the mystery of human life”.152 68. Applying Finnis’s theory to the present question, Finnis would likely suggest that paragraphs 8a to 8f should not be legally permitted, because such “private” gratification does not constitute “aesthetic experience” and “play”, such erosion of “human dignity” destroys “friendship” and one’s “conscience”, and such permissive laws detract from the “common good” which is a framework set of conditions that enables individuals to pursue human flourishing. This is so because, inter alia, such gratification is disconnected from the reality of sex that is a real interpersonal union and gift of mutual self-giving, and alienates people into a world of phantasy by fostering an illusion of intimacy through parasocial relationships.153 Moreover, although one’s conscience may be genuinely misinformed as to believe paragraphs 8a to 8f to be morally permissible, each individual is also responsible for forming his/her conscience well through education. 154 Overall, the erosion of Finnis’s basic goods and detractions from his principles of practical reasonableness are exacerbated when one generates non-consensual sexual deepfakes which objectifies and degrades human beings, and treats humans merely as a means to an end instead of an end in ourselves. D. Virtue Ethics 69. While there are numerous virtue ethics philosophers, such as Aristotle, Aquinas, Anscombe, MacIntyre, Pinckaers, etc, I will focus on the more traditional accounts of virtue jurisprudence, which place emphasis on telos,155 eudaimonia,156 human flourishing,

152

Planned Parenthood v Casey 505 U.S. 833 at 851 (1992). John Finnis, supra n 149; John Finnis, The Collected Essays of John Finnis: Volumes I-V (Oxford University Press, 2013); Pope John Paul II, Man and Woman He Created Them: A Theology of the Body (Pauline Books & Media, 2006); Karol Wojtyla, supra n 153; Edward Sri, Men, Women, and the Mystery of Love (Servant, 2007); Andrew Koppelman, “Is Pornography “Speech”?” (2008) Legal Theory 14(1) 71-89; John Finnis, “Reason and Passion: The Constitutional Dialectic of Free Speech and Obscenity” (1967) University of Pennsylvania Law Review 222. 154 Catechism of the Catholic Church (Vatican City: Libreria Editrice Vaticana, 2nd Ed, 1997) (“CCC”) § 17831785, 1791 and 1798. 155 Telos is the ancient Greek term for an end, fulfilment, completion, goal or aim; it is the source of the modern word “teleology”. Tad Brennan, “Telos” (2002) Routledge Encyclopedia of Philosophy. 156 Eudaimonia is the ancient Greek term for human flourishing, well-being and living an objectively good life. C.C.W Taylor, “Eudaimonia” (1998) Routledge Encyclopedia of Philosophy. 153

159

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

practical reason and wisdom, human dignity and the common good, to flesh out five key ideas that I believe to be particularly relevant to the question at hand. 70. First, unlike René Descartes who endorsed the dualism of body and soul, by suggesting, inter alia, that one’s body is something that one “has” and not “am” because “I think, therefore I am”, virtue ethicists often endorse the union of body and soul by highlighting that “I am my body, and my body is not merely a tool that I have”. 157 Flowing from this union of body and soul, and applying virtue ethics to the present question, one can better appreciate how every immoral act of generation and gratification in paragraphs 8a to 8f harms the actor himself/herself, how every decision changes one’s self regardless of whether we are cognisant of it or desire it, and how the actor himself/herself is a “victim” even if there appears to be “no harm” in such cases of “private” gratification. 71. Second, virtue ethicists often recognise that morality of human acts is not purely dependent on consequences or utilitarian in nature, but rests on three “sources” — the object, the intention, and the circumstances 158 — where the end does not justify the means, such that a good intention does not justify an evil act, although a bad intention can make a good act evil.159 Moreover, although the circumstances “cannot change the moral quality of acts themselves”,160 they “contribute to increasing or diminishing the moral goodness or evil of

157

René Descartes, Discourse on the Method (SMK Books, 2018); René Descartes, Meditations on First Philosophy (Oxford University Press, 2008); Pope John Paul II, supra n 153; Karol Wojtyla, supra n 153; Edward Sri, supra n 153; Aristotle, De Anima (On the Soul) (Penguin Classics, 1987); St. Thomas Aquinas, The Summa Theologica (Benziger Bros ed, 1947) (“ST”) I, q. 76; Gertrude Elizabeth Margaret Anscombe & Peter Thomas Geach, Three Philosophers (Cornell University Press, 1950); Alasdair MacIntyre, Dependent Rational Animals: Why Human Beings Need the Virtues (Open Court, 2001); Servais Pinckaers, Passions and Virtue (The Catholic University of America Press, 2017). 158 CCC, supra n 154, at § 1757; Aristotle, The Nicomachean Ethics (Penguin Classics, 2004); Aquinas, supra n 157 at ST I-II, q.18-20; Gertrude Elizabeth Margaret Anscombe, Intention (Harvard University Press, 2nd Ed, 2000); Servais Pinckaers, The Sources of Christian Ethics (The Catholic University of America Press, 3rd Ed, 1985); Pope John Paul II, Veritatis Splendor (1993). 159 CCC, supra n 154, at § 1753; Aristotle, supra n 158; Aquinas, supra n 157, at ST I-II, q.18-20; Anscombe, supra n 158; Pinckaers, supra n 158; Pope John Paul II, supra n 158. 160 CCC, supra n 154, at § 1754; Aristotle, supra n 158; Aquinas, supra n 157, at ST I-II, q. 7; Anscombe, supra n 158; Servais Pinckaers, supra n 158; Pope John Paul II, supra n 158.

160

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

human acts”,161 and the doctrine of double effect may be valid. 162 Overall, the act and movement of one’s will are essential in evaluating the morality of one’s actions. 72. Third, virtue ethicists often note that virtues are habits that can be cultivated, and which dispose a person to choose the good more easily, delightfully and consistently. 163 As per the Catechism of the Catholic Church, “[a] virtue is an habitual and firm disposition to do the good. It allows the person not only to perform good acts, but to give the best of himself. The virtuous person tends toward the good with all his sensory and spiritual powers; he pursues the good and chooses it in concrete actions”.164 Applying virtue ethics to the present question, the desire to utilise AI to generate sexual content for one’s “private” gratification reveals the vice of lust, and a need to grow in the cardinal virtue of temperance165 and virtue of chastity to counter this vice. To elaborate, three points are notable. First, the vice of lust stems from disordered sexual desire, as it entails the use or objectification of humans and is the result of too little, rather than too much, sexual desire whereby one only impersonally wants “X” from another person instead of wanting the other person personally in totality.166 Second, the virtue of chastity is not a prudish rejection of anything that is vaguely sexual in nature, but the recognition of the “integrity of the person” and the “integrality of the gift of self”. 167 Third, the cardinal virtue of temperance suggests that one’s passions and appetites must be educated and ordered by reason, intellect and will, such that one has true freedom from self-mastery instead of being enslaved by his/her passions, such as the disordered desire to utilise AI chatbots to generate sexual

161

CCC, supra n 154, at § 1754; Aristotle, supra n 158; Aquinas, supra n 157, at ST I-II, q. 7; Anscombe, supra n 158; Servais Pinckaers, supra n 158; Pope John Paul II, supra n 158. 162 Aquinas, supra n 157, at ST I-II, q. 64, a. 7; CCC, supra n 154, at § 2263-2265; Anscombe, supra n 158; Servais Pinckaers, supra n 158; Pope John Paul II, supra n 158. 163 CCC, supra n 154, at § 1810; Aristotle, supra n 158; Aquinas, supra n 157, at ST I-II, q. 55; Alasdair MacIntyre, After Virtue: A Study in Moral Theory (University of Notre Dame Press, 3rd Ed, 2007); Servais Pinckaers, supra n 158. 164 CCC, supra n 154, at § 1803. 165 CCC, supra n 154, at § 1805 and 1806. There are four cardinal virtues — prudence, fortitude, justice and temperance, and prudence is the “queen” of all virtues. 166 CCC, supra n 154, at § 2351; Pope John Paul II, supra n 153; Karol Wojtyla, supra n 153; Edward Sri, supra n 153; Robert W. Finn, “Blessed Are the Pure in Heart: A Pastoral Letter on the Dignity of the Human Person and the Dangers of Pornography” (2007) The Linacre Quarterly 74(4). 167 CCC, supra n 154, at § 2337 and 2346; John Paul II, supra n 153; Karol Wojtyla, supra n 153; Edward Sri, supra n 153.

161

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

content for one’s gratification. 168 In essence, the more virtuous one is, the more free one truly is, which leads me on to the next point. 73. Fourth, "the more one does what is good, the freer one becomes”. 169 Virtue ethicists often recognise that morality is not stifling or suffocating, or purely associated with prohibitions and obligations, but is truly freeing and liberating in Pinckaers’s sense of “freedom for excellence”, instead of “freedom of indifference”, as fleshed out through his piano analogy.170 As per Pinckaers, while “anyone is free to bang out notes haphazardly on the piano, as the fancy strikes him”, “the person who really possesses the art of playing the piano has acquired a new freedom” 171 — while the “freedom of indifference” is a “rudimentary, savage sort of freedom”, the “freedom for excellence" enables one to “compose new pieces” and “execute works of his choice with perfection”. 172 Applying virtue ethics to the present question, true freedom respects human dignity, where one has greater “freedom for excellence” or “[t]he power to act freely with excellence and perfection”.173 74. Fifth, virtue ethicists would likely agree with Pope Leo XVI’s comment in his first encyclical on AI, which drew inspiration from and built upon Pope Leo XIII’s 1891 encyclical on capital and labour, 174 that “the principles of Social Doctrine — the dignity of the person, the common good, the universal destination of goods, subsidiarity, solidarity and justice — [...] are criteria for judging whether technologies truly serve humanity or are subjugating it” which should serve as “guidelines for our decision-making” in the development of AI.175 Moreover, Pope Leo XVI noted that “‘[e]very human person possesses an infinite dignity, inalienably grounded in his or her very being, which prevails

168

CCC, supra n 154, at § 1804; Pope John Paul II, supra n 153; Karol Wojtyla, supra n 153; Edward Sri, supra n 153. 169 CCC, supra n 154, at § 1733. 170 Servais Pinckaers, supra n 158, at pp 354, 355 and 360. 171 Id, at p 355. 172 Id, at p 355. 173 Id, at p 375. 174 Pope Leo XIII, Rerum Novarum (1891). 175 Pope Leo XVI, Magnifica Humanitas (2026) at [183].

162

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

in and beyond every circumstance, state, or situation the person may ever encounter’— in other words, always and without exception”. 176 75. Overall, virtue ethicists may argue that AI chatbots should not be legally permitted to generate adult sexual content for adults’ “private” gratification, and will likely reject the concept of “adult sexual content” entirely as such content are also unsuitable for adults insofar as they do not make adults more virtuous people, because virtues are both aspirational and practical in nature — virtue ethicists are concerned with how one can become the most virtuous person practically possible, and not how “unvirtuous” one can be to still be considered “virtuous” overall. IV. OVERALL RECOMMENDATION: AI CHATBOTS SHOULD NOT BE LEGALLY PERMITTED TO GENERATE SEXUAL CONTENT FOR ADULTS’ “PRIVATE” GRATIFICATION 76. Overall, considering Finnis’s theory, virtue ethics, and various practical factors, I argue that AI chatbots should not be legally permitted to generate sexual content for adults’ “private” gratification. However, the application of Finnis’s theory and virtue ethics does not necessitate a one-size-fits-all regulatory response, for instance the criminalisation of paragraphs 8a to 8f, because the principle of subsidiarity is also applicable such that the question of how exactly each jurisdiction should regulate this issue is ultimately a jurisdiction-specific one.177 Nevertheless, introducing ex ante, instead of purely ex post, obligations will likely be vital to address this issue; and ensuring that online safety laws also cover standalone AI chatbots, in addition to embedded AI chatbots, will likely be critical too. At this juncture, I note that not all laws need necessarily be proactively enforced — for instance, sometimes the “naming-and-shaming” by regulators or “moral signalling effect” of laws may be equally effective, if not more effective, tools than formal legal enforcement in tackling online harms.178

176

Id, at [53]. Aquinas, supra n 157, at ST I-II, q. 96, a. 2. 178 Sharon Yadin, “Should Regulators Shame Companies into Compliance?” (18 September 2019) <https://www.theregreview.org/2019/09/18/yadin-should-regulators-shame-companies-into-compliance/> (accessed 13 June 2026); Sharon Yadin, “Regulatory Shaming” (2019) 49 Environmental Law 407; Cass R. Sunstein, “On the 177

163

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

77. Numerous practical factors also support the argument that AI chatbots should not be legally permitted to generate sexual content for adults’ “private” gratification. a. First, many jurisdictions do not impose age assurance requirements on AI chatbots, and even if they do, or if AI chatbot developers voluntarily self-declare to implement age assurance mechanisms,179 age assurance implementation is often ineffective due to multiple shortcomings such as the reliance on age self-declaration which one can simply lie through, the inaccuracy of age estimation and inference technologies (e.g. facial age estimation, behavioural signals, etc), and easy practical workarounds or circumventions including the use of Virtual Private Networks (“VPN”), misusing another’s identification details for age verification (e.g. ID, credit card, passport, etc), using web browsers instead of apps to access certain services, creating multiple accounts or shared accounts, borrowing another’s device to access certain services, manipulating parental control settings, etc. 180

Expressive Function of Law” (1996) 144 University of Pennsylvania Law Review; Yuval Feldman, “The Expressive Function of the Law: Legality, Cost, Intrinsic Motivation and Consensus” (2006) Bar-Ilan University Public Law & Legal Theory Research Paper Series. 179 Reuters, “OpenAI rolls out age prediction on ChatGPT” (21 January 2026) <https://www.reuters.com/technology/openai-rolls-out-age-prediction-chatgpt-2026-01-20/> (accessed 13 June 2026). 180 Simon Thorne, “ChatGPT is about to get erotic, but can OpenAI really keep it adults‑only?” (21 October 2025) <https://theconversation.com/chatgpt-is-about-to-get-erotic-but-can-openai-really-keep-it-adults-only-267660> (accessed 13 June 2026); United Kingdom Government, “Growing up in the online world: a national consultation” (2 March 2026) <https://www.gov.uk/government/consultations/growing-up-in-the-online-world-a-nationalconsultation> (accessed 13 June 2026); OECD, “Age assurance practices of 50 online services used by children” (25 June 2025) <https://www.oecd.org/en/publications/age-assurance-practices-of-50-online-services-used-bychildren_a19853ab-en.html> (accessed 13 June 2026); Transparency Coalition, “New survey finds overwhelming 96% support for protecting kids online” (10 February 2026) <https://www.transparencycoalition.ai/news/newsurvey-finds-overwhelming-support-for-protecting-kids-online> (accessed 13 June 2026); Wouter Lueks et al, “Assessing Age Assurance Technologies: Effectiveness, Side-Effects, and Acceptance” (26 March 2026) <https://arxiv.org/pdf/2603.25695> (accessed 13 June 2026); Sarah Koh, “Counter AI chatbot abuse by mandating user feedback channels, transparency: Experts” (2026) The Straits Times; Fatmaelzahraa Eltaher et al, “The Digital Loophole: Evaluating the Effectiveness of Child Age Verification Methods on Social Media” (2025); eSafety Commissioner, “Social Media Minimum Age: Compliance update” (March 2026) <https://www.esafety.gov.au/sites/default/files/202603/SocialMediaMinimumAgeComplianceUpdateMarch2026.pdf?v=1774905032806> (accessed 13 June 2026); eSafety Commissioner, Social Media Minimum Age Regulatory Guidance (September 2025) <https://www.esafety.gov.au/sites/default/files/2025-09/eSafety-SMMA-RegulatoryGuidance.pdf?v=1775448561246> (accessed 13 June 2026); Ofcom & Information Commissioner’s Office, “Age Assurance” (25 March 2026) <https://www.ofcom.org.uk/siteassets/resources/documents/online-safety/informationfor-industry/other/ofcom-ico-joint-statement-on-age-assurance.pdf?v=414966> (accessed 13 June 2026); Lim Sun Sun, “Minors’ online safety: The options beyond Australia’s social media ban” (2026) The Straits Times.

164

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

b. Second, if age assurance is not implemented effectively, and children can easily access AI chatbots or bypass ineffective age assurance mechanisms, all AI chatbots across the board should not be legally permitted to generate sexual content for both child and adult users to protect children. c. Third, research has shown that allowing AI chatbots to produce such sexual content disproportionately harms women, girls and children. 181 d. Fourth, allowing AI chatbots to produce such sexual content enables AI chatbot developers to profiteer off non-consensual sexual content and the vice of lust by offering “paid subscriptions”, where premium subscribers can pay to unlock more paywalled features to create more highly sexualised AI chatbot-generated content.182 This was already evident in both the Grok fiasco on X in early 2026 183 and the China Alien Chat case.184 78. At this juncture, it is notable that numerous non-exhaustive factors explain why traditional content regulation may occasionally be insufficient to address the challenges posed by AI chatbots. First, there is greater ease of misusing AI chatbots, than traditional media, to generate, publish and disseminate “non-consensual” sexual deepfakes. Second, there is greater ease of using AI chatbots, than traditional media, to generate sexual content in multimodal formats, such as images, videos, texts and audio. Third, the traditional “notice and takedown” reporting mechanisms on social media platforms will unlikely be completely transposable to the context of reporting mechanisms on AI chatbots, due to the “fleeting” or "temporal" nature of the content generated by AI chatbots, and the fact that AI chatbot responses are continually being iterated from multiple sources including model training, human feedback, user input, etc, such that “user feedback” mechanisms may be more appropriate for AI chatbots. 185

181

United Kingdom Government, supra n 110; Ysa Marie, supra n 5; Furizal et al, supra n 110; Meghan Donevan, supra n 110; Daniel Thomas, supra n 110. 182 Ashley Belanger, supra n 132; Ysa Marie, supra n 5. 183 Matt Burgess, supra n 132; Hadas Gold, supra n 132. 184 Alien Chat, supra n 36; Mary Bennett & Rob Robinson, supra n 36; Xiao Xiao, supra n 36. 185 Sarah Koh, supra n 180.

165

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

V. CONCLUSION 79. To conclude, I recommend that AI chatbots should not be legally permitted to generate sexual content for adults’ “private” gratification. 80. Nevertheless, I note that this is an ever-evolving field with numerous questions remaining unanswered. For example, how should liability be apportioned and attributed to different actors across the value chain — the LLM and AI chatbot developers, deployers, users, etc?186 Will the rise of “autonomous” agentic AI affect our answer to this question? 187 And relatedly, should criminal sanctions, regulatory and civil sanctions, or market selfregulation be imposed on or encouraged amongst such actors? Moreover, could the AI chatbot itself, instead of the content that it generates, constitute “sexual content” instead, especially in view of the rise of AI companions and nudification apps which tend to entail roleplay avatars or characters with a range of customisable features such as appearances, personalities, backstories, communication styles, etc? Lastly, situations in which the offplatform links generated by AI chatbots lead to paywalled or non-paywalled sexual content have not been discussed in depth. 81. To round off, I note that there is utility in investing in both ex ante and ex post nonregulatory tools and levers, in addition to legal levers, because the law alone cannot fully address the complex and multi-faceted problem of AI-related online harms. Examples of such non-regulatory tools and levers may include the development of ethical LLM and AI

186

Tech For Good Institute, “Criminalising Offensive Speech Made by AI Chatbots in Singapore” (14 November 2023) <https://techforgoodinstitute.org/insights/perspectives/criminalising-offensive-speech-made-by-ai-chatbotsin-singapore/> (accessed 13 June 2026); Josh Lee, “Criminalising Offensive Speech Made by AI Chatbots in Singapore” (22 August 2022) <https://lawtech.asia/criminalising-offensive-speech-made-by-ai-chatbots-insingapore/> (accessed 13 June 2026); Singapore Academy of Law Law Reform Committee, “Report on Criminal Liability, Robotics and AI Systems” (February 2021) <https://sal.org.sg/wp-content/uploads/2025/03/2021-Reporton-Criminal-Liability-Robotics-AI-Systems.pdf> (accessed 13 June 2026); Cheryl Seah, “Liability for AI-generated Content” (March 2024) <https://lawgazette.com.sg/feature/liability-for-ai-generated-content/> (accessed 13 June 2026); Frowe & Parry on NCII, supra n 100; Lawtech UK, “Public consultation – Liability for AI Harms under the private law of England and Wales” (January 2026) <https://lawtechuk.io/ukjt/public-consultation-liability-for-aiharms-under-the-private-law-of-england-and-wales/> (accessed 13 June 2026). 187 Clare McGlynn et al, supra n 13, at pp 29, 34, 43-44, 91; Infcomm Media Development Authority, “Discussion Paper: Legal Responsibility for AI Agents” (May 2026) <https://www.imda.gov.sg//media/imda/files/about/emerging-tech-and-research/artificial-intelligence/agents-legal-responsibility.pdf> (accessed 13 June 2026); Infcomm Media Development Authority, “Model AI Governance Framework for Agentic AI” (20 May 2026) <https://www.imda.gov.sg/-/media/imda/files/about/emerging-tech-and-research/artificialintelligence/mgf-for-agentic-ai.pdf> (accessed 13 June 2026).

166

Singapore Comparative Law Review

Should AI Chatbots Be Legally Permitted To Generate Sexual Content For Adults’ “Private” Gratification

chatbot developer, deployer and user guidelines or best practices, the rollout of educational online safety programmes and initiatives that also elaborate on virtue ethics, etc. 188 After all, as per Pope Leo XVI in his first encyclical on AI, “shared responsibility” is vital — AI must be co-developed “by institutions capable of regulating without stifling, and protecting without taking over; by businesses that recognize work and dignity as measures of success; by intermediary organizations and educational communities that rebuild trust and relationships; and by citizens who cultivate responsibility, moderation, discernment and a sense of truth”.189 Only in this way can “innovation genuinely serve integral human development, rather than becoming a source of exclusion and dominance”, and “the promise of progress be recognized as authentic, because it is measured against the inviolable dignity of every man and woman”.190

188

Zeve Sanderson & J. Scott Babwah Brennen, “When Age Assurance Laws Meet Chatbots” (5 September 2025) <https://www.techpolicy.press/when-age-assurance-laws-meet-chatbots/> (accessed 13 June 2026); Vanessa Lim, “'Nutrition labels' for AI apps among measures being studied to boost online safety: Josephine Teo” (31 March 2026) <https://www.channelnewsasia.com/singapore/ai-nutrition-labels-online-social-media-josephine-teo6021581> (accessed 13 June 2026); Infocomm Media Development Authority, “Model Al Governance Framework for Generative Al: Fostering a Trusted Ecosystem” (30 May 2024) <https://aiverifyfoundation.sg/wpcontent/uploads/2024/05/Model-AI-Governance-Framework-for-Generative-AI-May-2024-1-1.pdf> (accessed 13 June 2026); Infocomm Media Development Authority, “Starter Kit for Testing LLM-Based Applications for Safety and Reliability” (January 2026) <https://www.imda.gov.sg/-/media/imda/files/about/emerging-tech-andresearch/artificial-intelligence/large-language-model-starter-kit.pdf> (accessed 13 June 2026); Sugidha Nithiananthan & Racher Du, “Rethinking online safety in the age of deepfakes and nudifiers” (2026) The Straits Times; Clare McGlynn et al, supra n 13; IBM, supra n 13. 189 Pope Leo XVI, supra n 175, at [181]. 190 Ibid.

167

Outstanding.

Leaders.

HSF Kramer Singapore office offers a dual qualification training programme, resulting in being qualified in Singapore as well as England & Wales. The dual-qualification training programme is offered jointly with our Formal Law Alliance partner, Prolegis LLC. Training Contracts will be issued in accordance to the SILE moratorium time line. HSF Kramer Hong Kong office offers a Hong Kong training contract. We recruit 2 years in advance and future trainees are required to complete conversion exams and PCLL before starting the training contract. Trainee Solicitors are usually recruited via our vacation scheme programme.

The vacation scheme: Our month-long, paid vacation schemes are a chance to get to know us and the first step in starting a legal career with us. You will undertake real work in two different departments for high-profile clients; attend workshops, presentations and social events; and complete an exit interview with partners.

We are looking for exceptional and all-rounded candidates who are penultimate or final-year law students, and while we look for strong academic performers, we will always take mitigating circumstances into account – please include as much detail as you’re comfortable with in the additional information box on the online application form. If your initial application is shortlisted, you will be invited to undergo an online assessment. You will then be interviewed by a panel including a partner and a senior associate.

Applications for our Singapore and Hong Kong 2027 vacation schemes open from 1 September 2026 to 1 January 2027. Apply Now Hong Kong:

Apply Now Singapore:

Stay connected: @hsfkramerearlycareersapac

CAREERS.HSFKRAMER.COM/GLOBAL/EN/SINGAPORE/EARLY-CAREERS/OPPORTUNITIES CAREERS.HSFKRAMER.COM/GLOBAL/EN/HONG-KONG/EARLY-CAREERS/OPPORTUNITIES

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

STATUTORY SUTURES FOR THE LIABILITY GAP: A 'NO-FAULT' PRESCRIPTION FOR BLACK-BOX AI IN SINGAPORE HEALTHCARE Jolyn Ho Min Yee* ABSTRACT Black-box AI is increasingly prevalent in Singapore's clinical pathways, and its opacity strains liability premised on proof of fault. This article explores how Singapore's tort system and New Zealand's no-fault Accident Compensation Act 2001 compensate patients and deter clinicians, hospitals, and developers when black-box AI errors emerge. Singapore's regime holds while the clinician can check the machine, but opacity shields hospitals and developers from proof of breach and causation. New Zealand removes fault in form, yet negligence re-enters it through its statutory exclusions and high-risk patients are left unprotected. The article proposes a sector-specific nofault scheme for Singapore that harnesses the strengths of New Zealand's regime, equipped with a sunset clause that returns the field to tort once black-box AI is no longer opaque. I.

INTRODUCTION

With artificial intelligence (“AI”) increasingly incorporated in healthcare delivery in Singapore, patients enjoy quicker diagnosis and improved treatment outcomes. However, a specific class of AI, black-box AI, potentially introduces a novel liability problem due to its opaque reasoning. Singapore's current framework relies on the Bolam-Bolitho negligence standard and the expectation that clinicians will exercise meaningful "Human-in-the-Loop" (“HITL”) supervision over AI outputs. This article argues that for black-box deep learning, this framework becomes strained. The article explores the functionalist comparison between Singapore’s tort law and New Zealand’s no-fault Accident Compensation Act 2001 (“ACA”).1 Ultimately, it proposes a sector-specific no-

Singapore Management University, Yong Pung How School of Law, LLB Class of 2027. I am profoundly grateful to the editorial team for their support and feedback throughout the drafting process. All errors that remain are solely my own. 1

Accident Compensation Act 2001 (New Zealand).

168

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

fault compensation scheme incorporating a triage mechanism. This scheme seeks to overcome the evidentiary difficulties patients face when seeking compensation for injuries involving black-box AI errors. It aims to achieve this while preserving the deterrence function to ensure that clinicians, hospitals, and developers are responsible for patient safety. II.

WHAT IS BLACK-BOX AI IN HEALTHCARE? A. Mechanism and Opacity of Black-Box AI

Deep learning systems are a form of AI, where developers do not program the model to reflect their understanding of the causal relationship between the inputs and outputs. 2 Instead, these systems have multiple nested layers of interconnected nodes which activate when they detect features in the input data and learn how to recognise patterns. 3 The advantage of such AI systems is that they can be trained using millions of inputs to achieve extremely precise predictions. 4 However, these AI systems are ‘black-boxes’ in that they exhibit training and inference opacity. Training opacity arises as these systems are so highly sensitive to minute dataset alterations that developers cannot predict the final model based purely on its inputs.5 Hence, even the original experts who built the system cannot inspect the final, trained AI and explain the mathematical reasoning behind the generation of the outputs. Inference opacity refers to the idea that the AI system relies on a colossal number of parameters tied to features it has learned during training to generate outputs.6 As a result, the connections it draws exceed the limits of our working human memory and examining a diagnostic AI system will not aid a clinician in understanding the model’s reasoning.7

Alex John London, 'Artificial Intelligence and Black-Box Medical Decisions: Accuracy versus Explainability' (2019) 49(1) Hastings Center Rep 15, 16. 3 London (n 2) 16. 4 London (n 2) 16.London at 16. 5 Joshua Hatherley, Lauritz Aastrup Munch and Jens Christian Bjerring, 'In Defense of Post Hoc Explanations in Medical AI' (2026) 56 Hastings Center Report 40, 42. 6 Hatherley, Munch and Bjerring (n 5) 42. 7 Hatherley, Munch and Bjerring (n 5) 42.

169

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

B. Prevalence in Singapore Despite this opacity, Singapore’s healthcare clusters embed black-box AI and machine learning models in diagnostic and treatment pathways to improve healthcare delivery. The ‘Singapore Eye Lesion Analyser’ (“SELENA+”) is an AI system that evaluates retinal photographs to screen for diabetic eye disease, glaucoma, and age-related macular degeneration. 8 SELENA+ assesses retinal photographs faster while maintaining the same level of accuracy as human graders. 9 As the AI reduces the need for ophthalmologists to screen every case, 10 it scales up population-level screening cost-effectively. This facilitates early clinical intervention, in turn preventing blindness in many patients. Other black-box AI tools are similarly integrated into institutional workflows because they offer faster diagnosis, reduced specialist workload, and improved screening sensitivity. AimSG triages chest radiographs in emergency care to flag urgent cases, and FathomX seeks to replace the second radiologist reader in screening mammography. 11 C. Risk of Patient Harm For all their benefits, black-box AI systems are not infallible. The AI in Healthcare Guidelines 2.0 (“AIHGle”),12 which was co-developed by the Ministry of Health, the Health Sciences Authority and Synapxe, outlines how AI can compromise patient care outcomes. The risks emerge at different stages of the AI lifecycle, such as development risk from poor training data quality or bias, deployment risk in insufficient monitoring and audit mechanisms, and risk from misinterpretation and inappropriate use of AI by clinical staff. 13 Poor training data quality and insufficient monitoring by developers and hospitals respectively result in patient harm as black-box models suffer from model drift. 14 This is a phenomenon where 8

Warren B Chik and Haran Sugumaran, 'Legal governance of artificial intelligence in healthcare devices and services in Singapore' in Barry Solaiman and I Glenn Cohen (eds), Research Handbook on Health, AI and the Law (Edward Elgar Publishing 2024) 390. 9 Chik and Sugumaran (n 8) 391. 10 Tan Cher Heng and Benjamin Seet, 'How AI Will Change Medicine' SMA News (October 2023) 16. 11 Tan and Seet (n 10) 16. 12 Ministry of Health, Health Sciences Authority and Synapxe, 'Artificial Intelligence in Healthcare Guidelines Version 2.0' (March 2026). 13 Ministry of Health, Health Sciences Authority and Synapxe (n 12) 5. 14 Ministry of Health, Health Sciences Authority and Synapxe (n 12) 5.

170

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

the model's performance degrades over time because the real-world data it encounters begins to change from the data it was initially trained on. 15 As a locked model cannot adapt on its own, its predictions degrade as local disease patterns or demographics shift away from the training data, unless the hospital detects the drift through monitoring. The AIHGle also warns of hallucinations, where AI models generate incomplete or entirely fictitious clinical content presented confidently as fact. 16 The opacity of black-box systems complicates the process for clinicians. If an AI model hallucinates a misdiagnosis and the doctor cannot verify the AI's internal logic to catch the error, the fictitious output may be relied upon, resulting in delayed treatment or death. On the other hand, where the doctor can check the AI output against other tests, blindly relying on the AI would constitute inappropriate use. This is because AI should not replace professional judgement.17 The danger of scalable errors compounds the issue, because AI errors can be far more widespread than a single clinician's mistakes. 18 When a single clinician makes a mistake, the harm is generally isolated to one patient. However, consider the scenario where a flawed algorithmic rule is embedded within a black-box AI system that is deployed across Singapore's major public healthcare clusters. As the opacity makes the algorithmic error difficult to trace, 19 that single hidden error can lead to doctors misdiagnosing or mistreating a massive number of patients simultaneously before the root cause is identified. Given the opacity of black-box AI that prevents clinicians from understanding the internal reasoning of these systems, this article will examine in what situations the standard of care and causation would be strained in medical negligence.

Keyvan Rahmani and others, 'Assessing the Effects of Data Drift on the Performance of Machine Learning Models Used in Clinical Sepsis Prediction' (2023) 173 Intl J Med Informatics 104930, 2. 16 Ministry of Health, Health Sciences Authority and Synapxe (n 12) 34. 17 Ministry of Health, Health Sciences Authority and Synapxe (n 12) 30. 18 Gary Kok Yew Chan, 'Medical AI, Standard of Care in Negligence and Tort Law' in Gary Chan and Yip Man (eds), AI, Data and Private Law: Translating Theory into Practice (Hart Publishing 2021) 175. 19 Cheryl Seah, 'Liability Arising from the Use of Artificial Intelligence' (The Singapore Law Gazette, May 2023) section 1.2 https://lawgazette.com.sg/feature/liability-arising-from-the-use-of-artificial-intelligence/ accessed 24 July 2026.

171

Singapore Comparative Law Review

III.

Statutory Sutures For The Liability Gap

THE COMPARATIVE DESIGN: A SEPSIS PREDICTION SCENARIO

A functionalist comparative law lens entails that law is understood in light of its functional relation to society.20 The function acts as the basis of comparison, and doctrinally different legal systems are comparable if they fulfil similar functions. 21 In this comparative design, the basis of comparison is how Singapore and New Zealand functionally address the societal problem of blackbox AI medical errors through patient compensation and deterrence. The functional analysis would be examined through a scenario involving the Sepsis Early Risk Assessment (“SERA”) algorithm. As an EHR-based prediction algorithm, SERA is a black-box AI system developed to predict sepsis up to 48 hours before onset. 22 Sepsis is a life-threatening medical emergency triggered by the body's extreme inflammatory response to an infection. 23 If left untreated, this severe reaction can impair blood flow to vital organs, ultimately leading to tissue damage, multi-organ failure, and death.24 Testing has shown that SERA significantly outperforms human physicians in predicting sepsis, enabling early intervention and better survival rates. 25 It achieves this through processing 115 variables drawn from electronic health records, such as lab results and clinicians’ notes. 26 SERA is inference-opaque, as no clinician can trace how 115 variables combine into a risk score. More important for liability, its output has no referent against which it can be checked. A SELENA+ grade can be verified against the retinal photograph itself. SERA's prediction can only be tested against the patient's observable symptoms and corroborative tests, which may not yet exist for a condition predicted up to 48 hours before onset. The hypothetical will explore and compare how Singapore and New Zealand’s compensation and deterrence functions operate where SERA fails to flag a patient who goes on to develop sepsis. The developer trained it on poor quality

Ralf Michaels, 'The Functional Method of Comparative Law' in Mathias Reimann and Reinhard Zimmermann (eds), The Oxford Handbook of Comparative Law (OUP 2006) 342. 21 Michaels (n 20) 342. 22 Kim Huat Goh and others, 'Artificial Intelligence in Sepsis Early Prediction and Diagnosis Using Unstructured Data in Healthcare' (2021) 12 Nature Communications 711, 2. 23 Candice Cai, 'What is sepsis?' (Singapore General Hospital, 12 October 2023) https://www.sgh.com.sg/news/singapore-health/what-is-sepsis accessed 26 July 2026. 24 Cai (n 23). 25 Goh and others (n 22) 7. 26 Goh and others (n 22) 3, 8.

172

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

data and neither developer nor hospital adequately validated and monitored it. Thereafter, treatment was delayed and the patient suffered organ damage. IV.

THE SCENARIO IN SINGAPORE LAW

In this scenario, the traditional tort concepts of fault and causation are strained, mainly regarding the regulation of developers and hospitals. A. Clinicians: Liability Turns on Corroborating the AI Even when faced with the inference opacity of SERA and an erroneous output, doctors need to reasonably support their eventual findings to avoid liability for medical negligence. As held in Khoo James v Gunapathy d/o Muniandy (“Khoo James”), to determine if a clinician is liable for medical negligence, Singapore applies the Bolam-Bolitho framework.27 Under the Bolam test, a medical practitioner is not considered negligent if their actions accord with a practice accepted as proper by a responsible body of medical opinion, even if other practitioners might take a contrary view.28 The Bolitho addendum establishes that this medical opinion must also have a logical basis.29 The judge would have to be satisfied that the experts: (a) directed their minds to the question of comparative risks and benefits; and (b) reached a defensible conclusion. 30 Even if the use of SERA is a practice accepted as proper by a responsible body of medical opinion as it is well integrated in clinical workflows, inference opacity poses difficulties for medical experts to satisfy the Bolitho addendum’s threshold of logic. 31 The Court of Appeal discussed the application of the Bolitho addendum in Khoo James.32 Medical experts who make bare and unsupported assertions would fail limb (a), which requires them to direct their minds to the question of comparative risks and benefits. 33 In limb (b), a defensible conclusion requires a

[2002] SGCA 25, [2002] 1 SLR(R) 1024 [64]. Bolam v Friern Hospital Management Committee [1957] 1 WLR 582 (QBD) 587. 29 Bolitho v City and Hackney Health Authority [1998] AC 232, 241–242. 30 Bolitho (n 29) 241–242. 31 Chan (n 18) 179. 32 (n 27) [63]–[69]. 33 Khoo James (n 27) [64]. 28

173

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

medical opinion that is “internally consistent on its face” such that it is not self contradictory. 34 The opinion must also not “fly in the face of proven extrinsic facts relevant to the matter” such as by ignoring known medical facts or advances in medical knowledge. 35 In Khoo James, the court found that interpreting complex brain scans was an inexact science. 36 The interpretive differences arose from multiple factors, such as the considerable expertise required to reconstruct a three-dimensional image of the brain from two-dimensional MRI slices and an inherent margin of error from the ill-defined borders of the nodule.37 Nonetheless, the experts’ diagnosis passed the threshold of logic. They had addressed their minds to the mechanics behind the interpretation of such scans, properly weighed the perceived growth against the tumour's previously benign histology, and there were no contradictory extrinsic medical facts. 38 In contrast, the mechanics behind the interpretation of SERA are shrouded by inference opacity. This makes it difficult for clinicians to logically justify their assessments of comparative risks and benefits, complicating the direct application of the Bolam-Bolitho test.39 Professor Chan asserts that since doctors do not possess the expertise to interpret the reasoning of medical AI, courts should not be strictly bound to only the Bolam-Bolitho test to assess the reasonableness of the doctor’s conduct.40 Instead, courts should examine the standard factors such as risks and extent of harm and the costs of taking precautionary measures. 41 Professor Chan raises the example of Hii Chii Kok v Ooi Peng Jin London Lucien (“Hii Chii Kok”), where the court held that the doctor was not negligent for heavily relying on the high likelihood that a device’s positive result is truly positive.42 The Gallium scan was the most sensitive and advanced diagnostic tool at that time. 43 While there was a likelihood of inaccurate results, the results of the scan pointed in the malignant direction and the doctor conducted complementary

Khoo James (n 27) [65]. Khoo James (n 27) [65]. 36 (n 27) [81]. 37 Khoo James (n 27) [81]. 38 Khoo James (n 27) [94]. 39 Chan (n 18) 179. 40 Chan (n 18) 181. 41 Gary Chan Kok Yew, Health Law and Medical Ethics in Singapore (Routledge 2021) 277. 42 [2016] SGHC 21, [2016] 2 SLR 544 [162]; Chan (n 18) 183. 43 Hii Chii Kok (n 42) [162]. 35

174

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

tests that corroborated his diagnosis.44 Hence, doctors who use a black-box AI tool that has been proven to be highly accurate in clinical trials, and support their findings with observable symptoms and corroborative tests would not be in breach of their standard of care. 45 This approach accords with the recommendations of the AIHGle. As a regulatory guideline, the AIHGle can be used as an indicator of the industry expectations of AI use. 46 It provides that human oversight is mandatory for all clinical use of AI. 47 More specifically, it states that healthcare professionals should minimally be able to recognise limitations such as when the AI solution is not performing as expected. 48 Recognising limitations would entail conducting further investigations to check against SERA’s output. To establish causation under the “but-for” test, the court in Armstrong v Quest Laboratories Pte Ltd was not concerned with what the defendant ought to have done, but whether the defendant’s action or omission was a necessary condition for the plaintiff’s loss.49 In the SERA case, the court would ask whether a clinician's blind reliance on SERA was a necessary condition for the patient's untreated progression of sepsis. Had the clinician reviewed the other available patient data, they would have reached a different conclusion, and administered timely treatment. As that intervention would have prevented sepsis, the causation test would be satisfied. A benefit of this approach is that the tort system is flexible enough to assess the doctor’s standard of care, even with the inference opacity of black-box AI.50 Doctors are at risk of becoming liability sponges when they are strictly held accountable for any AI error resulting in patient harm, despite having no meaningful ability to affect the outcome. 51 This would be the case whereby the court requires doctors to defend their use of SERA by explaining its internal mechanisms, which is an insurmountable task. By acknowledging that the high accuracy of SERA and corroborative test 44

Hii Chii Kok (n 42) [162]–[166]. Chan (n 18) 183. 46 Gary KY Chan, 'AI in healthcare: Regulatory guidelines and judge-made negligence principles for AI implementers' (2025) Medical Law International 144, 156. 47 Ministry of Health, Health Sciences Authority and Synapxe (n 12) 11. 48 Ministry of Health, Health Sciences Authority and Synapxe (n 12) 31. 49 [2019] SGCA 75, [2020] 1 SLR 133 [77]. 50 Chan (n 18) 197. 51 Rebecca Crootof, Margot E Kaminski and W Nicholson Price II, 'Humans in the Loop' (2023) 76 Vand L Rev 429, 483, citing MC Elish and Tim Hwang, 'Praise the Machine! Punish the Human! The Contradictory History of Accountability in Automated Aviation' (Working Paper, 2015) 15 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2720477. 45

175

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

results weighs in favour of reasonable use, the Hii Chii Kok principle avoids treating doctors as liability sponges. The standard of care would account for the information the doctor is reasonably relying on. It excludes liability where doctors do not have technical expertise to exercise oversight, such as when SERA malfunctions due to poor training data and insufficient validation and monitoring. The standard of care would simultaneously deter automation bias. This is the phenomenon where clinicians over-rely on AI as a replacement for independent judgement. 52 Automation bias can manifest in patient harm from following incorrect advice and failing to act because the machine erroneously omitted to prompt for intervention. 53 Ultimately, where a doctor blindly relies on SERA, Singapore’s tort regime functions effectively to compensate the patient. It achieves this while promoting patient safety and deterring the errant doctor. B. Hospitals and Developers: Fault Provable in Principle, Unprovable in Practice Evidential difficulties in proving the Duty of Care of Hospitals and Developers Where the clinician has discharged their HITL role, however, the inquiry shifts from the clinician's substantive medical decision to whether the hospital and the AI developer have taken reasonable procedural steps in ensuring AI reliability, in line with industry practices set out in the AIHGle. 54 In Noor Azlin bte Abdul Rahman v Changi General Hospital Pte Ltd, the court held that the hospital owed a duty to ensure that it had in place a system which would allow for the proper management of each patient.55 This management includes providing doctors avenues for proper follow-up of radiological results, by allowing for consolidation of patient information and a procedure for recording medical decisions.56 In this scenario, the hospital’s duty is extended to a system of management for black-box AI devices, where the hospital is capable of acting in accordance with

Kate Goddard, Abdul Roudsari and Jeremy C Wyatt, 'Automation bias: a systematic review of frequency, effect mediators, and mitigators' (2012) 19 J Am Med Inform Assoc 121, 121, citing K Mosier and LJ Skitka, 'Human decision makers and automated decision aids: made for each other?' in R Parasuraman and M Mouloua (eds), Automation and Human Performance: Theory and Applications (Lawrence Erlbaum Associates 1996). 53 Goddard, Roudsari and Wyatt (n 52) 121. 54 Chan (n 46) 159, 165. 55 [2019] SGCA 13, [2019] 1 SLR 834 [101]. 56 Noor Azlin (n 55) [99]–[101].

176

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

soft regulatory guidelines of the AIHGle. 57 Hence, a hospital would also have a duty to ensure adherence to procedural safeguards. To establish an AI developer’s duty of care towards patients, the manufacturer liability principle in Deutz Far East (Pte) Ltd v Pacific Navigation Co Pte Ltd (“Deutz”)58 can be integrated into the first stage of the Spandeck framework to satisfy the requirement of legal proximity. 59 Spandeck establishes a duty of care through a threshold of factual foreseeability, followed by sufficient legal proximity between the parties, and no negating policy considerations. 60 The court in Deutz held that a manufacturer owes a direct duty of care when a product is intended to reach the consumer without any reasonable likelihood of an intermediate examination that would reveal latent defects or dangers.61 Inference opacity precludes doctors from conducting a meaningful intermediate examination to uncover hidden algorithmic defects before clinical application. In contrast, the AI developer retains exclusive control over the training data and algorithmic architecture that generate the risks. While a hospital's adherence to regulatory procedural safeguards establishes baseline system reliability, it cannot serve as an intermediate safety filter for all case-specific algorithmic errors. This is because the hospital cannot lift the opacity of the system, 62 leaving the patient vulnerable. The absence of a viable intermediate safety filter satisfies the proximity requirement under Spandeck, bridging the gap between the independent AI developer and the patient to establish a prima facie duty of care. Moreover, promoting beneficial technological innovations in healthcare must be balanced against the need for patient safety. 63 Thus, the pursuit and implementation of new technology is unlikely to be a policy consideration negating the duty. Poor-quality training data thus constitutes a breach

Ministry of Health, Health Sciences Authority and Synapxe (n 12) 21–29; Chan (n 46) 156. [1989] SGHC 88, [1989] 2 SLR(R) 392. 59 Spandeck Engineering (S) Pte Ltd v Defence Science & Technology Agency [2007] SGCA 37, [2007] 4 SLR(R) 100 [73], [115]. 60 Spandeck (n 59) [73], [115]. 61 Deutz (n 58) [104]–[106]. 62 London (n 2) 17; W Nicholson Price II, Sara Gerke and I Glenn Cohen, 'Liability for Use of Artificial Intelligence in Medicine' in B Solaiman and IG Cohen (eds), Research Handbook on Health, AI and the Law (Edward Elgar Publishing Ltd 2024) 159. 63 Gobinathan Devathasan v Singapore Medical Council [2010] SGHC 51, [2010] 2 SLR 926 [62]. 58

177

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

of this duty, as developers are expected to take reasonable steps to prevent foreseeable harms arising from unrepresentative or flawed datasets. 64 However, patients face formidable difficulties in obtaining data to prove fault of the developer or hospital. As Singapore does not have AI-specific legislation,65 patients rely on traditional civil procedure rules in standard discovery applications for evidence gathering. During standard discovery applications, the owner or manufacturer of black-box AI devices will likely “object on the grounds of confidentiality and trade secrecy, or protecting intellectual property, to prevent the software from being the subject of scrutiny by an independent third party.” 66 Courts are also reluctant to grant such discovery requests, holding that the request is unnecessary and would incur extra costs.67 As such, it is a challenge for injured patients relying on standard discovery rules to compel their disclosure to prove a breach of the standard of care by developers or hospitals. Even after obtaining evidence, SERA’s algorithm is not easily explainable.68 Patients are saddled with the prohibitive costs of hiring AI expert witnesses to interpret the particular defect in the developer’s source code or training data, or the hospital’s failure to identify an error in a validation exercise. Difficulties in establishing Causation regarding Hospitals and Developers Assuming duty and breach can be established regarding the developer’s poor training data and the hospital’s insufficient validation and monitoring, causation is challenging to prove because of the opacity of black-box algorithms and the multiplicity of actors involved in the AI’s lifecycle. To prove "but-for" causation, a patient must show that without the developer's use of poor training data or the hospital's failure to properly validate and monitor the system, the injury would not have occurred. However, because deep-learning algorithms cannot provide causal explanations for their outputs in terms that are interpretable by humans, it is impossible to ascertain the precise data point

Chan (n 46) 165. Cheng Kit Pang, 'A Comparative Analysis of Artificial Intelligence Regulation: Implications for Singapore' (2025) 37 SACLJ 189, 190. 66 Daniel Seng and Stephen Mason, 'Artificial Intelligence and Evidence' (2021) 33 SACLJ 241, 273 67 Seng and Mason (n 66) 273. 68 Jonathan Ming En Tan, 'Non-Deterministic Artificial Intelligence Systems and the Future of the Law on Unilateral Mistakes in Singapore' (2022) 34 SACLJ 91, 105. 65

178

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

or algorithmic weight SERA relied upon to generate a flawed recommendation. 69 Consequently, patients face an evidentiary hurdle in definitively untangling whether a misdiagnosis was physically caused by the developer's unrepresentative training data, the hospital's failure to perform procedural verification checks, or the inherent unpredictability of SERA. Where scientific uncertainty makes it impossible to pinpoint the exact cause of the harm, courts have applied an alternative test allowing a plaintiff to succeed if the defendant's breach materially increased the risk of injury. 70 This was applied to medical negligence in Singapore in Surender Singh s/o Jagdish Singh v Li Man Kay, where a hospital was held liable because its failure to monitor a patient materially increased the risk of fatal complications, bypassing the need to prove exactly how the injury physically manifested. 71 Despite its flexibility, the material contribution to risk test could still fail under the limitations established in Wilsher v Essex Area Health Authority.72 In Wilsher, a premature infant was negligently administered excess oxygen, but there were four other completely different natural conditions that could have independently caused the infant's blindness.73 As the plaintiff could not prove which of these independent agents actually caused the damage, the court refused to apply the test. 74 This maps directly onto the "problem of many hands" in medical AI, because a wide range of parties, including developers, manufacturers, researchers, and clinicians, are involved in building and deploying a single system. 75 An AI error could independently originate from defective code, poor training data, the hospital’s failure to monitor, 76 or an unforeseeable failure of the model itself.77 Hence, a court applying Wilsher would conclude that these represent distinct, independent possible causes. The causation claim would fail if the plaintiff cannot definitively isolate which cause was the one that materially contributed to the risk of harm.

Chan (n 18) 186; Chan (n 46) 165. McGhee v National Coal Board [1973] 1 WLR 1 (HL) 6. 71 Surender Singh s/o Jagdish Singh And Another (administrators of the estate of Narindar Kaur d/o Sarwan Singh) v Li Man Kay and Others [2009] SGHC 168, [2010] 1 SLR 428 [240]. 72 Wilsher v Essex Area Health Authority [1988] AC 1074 (HL). 73 Wilsher (n 72) 1090-91. 74 Wilsher (n 72) 1091. 75 Helen Nissenbaum, 'Accountability in a Computerized Society' (1996) 2(1) Sci Eng Ethics 25, 28. 76 Chan (n 46) 165. 77 Seah (n 19) pt 2(1.3). 70

179

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

Furthermore, the developer or hospital may argue that the unpredictable algorithmic failure is tantamount to a supervening event that breaks the chain of causation. 78 The unpredictability is a fundamental characteristic of SERA, as even high-performing trained models can fail in unpredictable ways that its opacity prevents anyone from foreseeing. 79 In particular, dataset shift occurs when a disparity between a model's original training distribution and its post-deployment environment causes its accuracy to decline over time. 80 Hence, part of AI’s future behaviour is inherently outside of the control of its programmers and users. Such an argument would prevent the patient from establishing causation to hold the developer or hospital liable. In summary, even if duty of care and breach can be established against the hospital and developers for procedural breaches, the difficulties in proving causation absolve them of liability. Where a poor dataset or algorithmic flaw leads to a pattern of incorrect outputs that causes widespread harm, developers and hospitals could point towards the AI’s intrinsic unpredictability to sever the chain of causation. Consequently, the tort system undermines deterrence against irresponsible AI development and deployment.

On the other hand, holding hospitals and developers as absolute underwriters for all unforeseeable AI-induced risks is considered anathema to innovation and stifles the adoption of life-saving medical technologies.81 This makes the current tort system inadequate to compensate patients for procedural breaches committed by developers and hospitals, endangering patient safety when black-box AI is involved in healthcare delivery.

Chan (n 46) 165. Virginia Dignum, Responsible Artificial Intelligence: How to Develop and Use AI in a Responsible Way (Springer 2019) 28–29. 80 Joshua Hatherley, 'A Moving Target in AI-Assisted Decision-Making: Dataset Shift, Model Updating, and the Problem of Update Opacity' (2025) 27 Ethics Inf Technol 20, 2. 81 Noah Lim Fang-Zhou and Tan Kay Shin, ‘The New Frontier: Regulating Artificial Intelligence in Singapore’ (2025) 37 SACLJ 436, 453. 79

180

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

THE SCENARIO IN NEW ZEALAND LAW, COMPARED WITH THAT IN SINGAPORE LAW

In contrast to Singapore’s tort system where patients claim against developers, hospitals, or doctors, New Zealand’s Accident Compensation Corporation (“ACC”) administers a levy-funded scheme under the ACA. Compensation under New Zealand’s treatment injury scheme requires only a connection between the medical treatment and the injury. 82 While proving negligence is irrelevant, causation is required. The claim would be denied if the injury was an ordinary consequence of the treatment itself, 83 or if the injury is a necessary part of the treatment process. 84 The patient has to first demonstrate that the treatment was directed by a Registered Health Professional (“RHP”).85 As the human doctor chose to use the AI, the treatment was given "by, or at the direction of" an RHP.86 A. Establishing Treatment Injury in New Zealand is easier than Establishing Fault in Singapore Additionally, the event must qualify as “treatment”. Under section 33 of the ACA, "treatment" is given a wide definition. 87 It explicitly includes "a diagnosis of a person's medical condition" 88 and the failure of any equipment or device. 89 While there is no AI-specific case, in the New Zealand Court of Appeal case of McGougan v DePuy International Ltd, patients who suffered severe personal injuries because of defective hip implants were covered under the ACA.90 They satisfied the criteria under the ACA, as they suffered personal injury in New Zealand, which included physical injuries and damage to prostheses that replace a part of the human body, and arose from receiving treatment by at least one RHP.91 No proof of fault is involved. Similarly, if a black-box AI generates a dangerous 82

Albert Andrew, ‘A No-Fault Approach to Medical Error: A New Zealand Perspective’ (2025) 9 J Hosp Manag Health Policy 33, 34. 83 ACA, s 32(1)(c). 84 ACA, s 32(1)(c). 85 ACA, ss 32(1)(a)(i) and 32(1)(a)(ii). 86 Christopher Ryan Boniface, The Legal Impact of Artificial Intelligence on the New Zealand Health System (PhD thesis, University of Canterbury 2021) 203. 87 ACA, s 33; Stephen Todd, ‘Treatment Injury in New Zealand’ (2011) 86 Chi-Kent L Rev 1169, 1198. 88 ACA, s 33(1)(b). 89 ACA, s 33(1)(g). 90 [2018] NZCA 91 [1], [4]. 91 McGougan (n 90) [16].

181

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

misdiagnosis, this constitutes a failure of the device itself under s 33(1)(g), triggering coverage without requiring the claimant to prove flawed developer training data. Crucially, the ACA definition of treatment also includes the "application of any support systems, including policies, processes, practices, and administrative systems". 92 An injury produced by a hospital's reliance on a clinical AI system is therefore routed through section 33(1)(h) without any finding of institutional negligence. At this stage, the approach differs starkly from Singapore’s tort system, where claimants need to prove fault and expend significant time and money hiring AI experts and obtaining evidence. This is because the New Zealand no-fault scheme was specifically enacted to replace the lottery of common law claims that enabled few claimants to succeed. 93 The ACA espouses the goals of the Woodhouse Report.94 It noted that due to the requirement to prove fault, common law claims were described as a lottery that gave rise to erratic achievements and inconsistent solutions. 95 The solution to an unpredictable common law system for compensation was the no-fault scheme which channeled funds to victims, instead of other actors such as insurers and lawyers. 96 Therefore, New Zealand’s compensation scheme seems superior for patient compensation because claimants do not need to prove negligence against an identified professional, which eliminates the associated costs and time for compensation. B. Establishing Causation is easier but Difficulties emerge in overcoming Statutory Clauses Proving that a black-box AI malfunction caused the injury poses significant evidentiary hurdles due to inference opacity and the multiplicity of parties involved, as shown in the analysis above involving Singapore tort law. These evidentiary difficulties could be overcome by the New Zealand tests for causation, but statutory clauses restrict claims. Under New Zealand’s ACA, traditional tort inquiries into causation are modified to suit a framework that focuses on compensating patients rather than assigning blame. The principles of robust inferences, material 92

ACA, s 33(1)(h). McGougan (n 90) [26]. 94 McGougan (n 90) [26]. 95 McGougan (n 90) [26]; citing Royal Commission of Inquiry, Compensation for Personal Injury in New Zealand (Government Printer 1967) paras 1, 14. 96 McGougan (n 90 ) [26]; citing Royal Commission of Inquiry, Compensation for Personal Injury in New Zealand (Government Printer 1967) para 59. 93

182

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

contribution, and the rejection of intervening events were developed across the cases of ACC v Ambros97 and W v ACC.98 In ACC v Ambros, the core issue was whether the doctors’ omissions to diagnose and treat Mrs Ambros’s spontaneous coronary artery dissection (“SCAD”) caused her death, given the scientific uncertainty surrounding whether any treatment would have successfully saved her from this rare disease.99 The Court held that the legal burden to prove causation on the balance of probabilities remains with the claimant. 100 However, because medical science often cannot provide absolute certainty, the court mitigated this strict threshold by allowing decision-makers to draw robust inferences of causation in some cases of uncertainty. 101 If medical science evinces a possible connection, a Judge may decide that causation is probable after examining all evidence. 102 In W v ACC, there were multiple contributing causes to the claimant’s mental injury, one of the causes being a serious assault when the claimant was a baby.103 The Court held that in such cases, the decision-maker should ask if the claimant’s physical injuries materially contributed to the mental injuries that they suffered. 104 Unlike the law in Singapore, the Court held that novus actus interveniens is an inappropriate concept in the context of the no-fault principles underpinning the scheme.105 The doctrine relieves the defendant of liability due to policy reasons, but the ACA scheme endeavours to provide cover regardless of fault of the individual who may have caused the injury.106 While intervening events may be considered in issues of proximity, the proximity step bars cover for unrealistic claims in rare cases, where there is a tenuous link between the claimant’s physical and mental injuries. 107 Applying to the context of the SERA hypothetical, causation can be established with the above causation principles. The patient can overcome the issue of inference opacity. In ACC v Ambros,

[2007] NZCA 304, [2008] 1 NZLR 340. [2018] NZHC 937, [2018] 3 NZLR 859. 99 Ambros (n 97) [1], [104]–[105]. 100 Ambros (n 97) [1], [65]. 101 Ambros (n 97) [67]. 102 Ambros (n 97) [68]. 103 W (n 98) [68]. 104 W (n 98) [68]. 105 W (n 98) [71]. 106 W (n 98) [71]. 107 W (n 98) [72], [75]. 98

183

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

the medical experts could not say that earlier diagnosis and treatment would, more probably than not, have averted death from SCAD, a rare condition with an unpredictable course. 108 Similarly, experts in SERA’s case do not need to prove the exact algorithmic failure. The patient can ask the decision-maker to draw a "robust inference" of causation. As long as experts concede there is a possible connection between the use of the AI and the misdiagnosis, the court can look at the totality of the circumstances, such as the temporal proximity of SERA’s output to the doctor's delayed intervention, or statistical evidence regarding SERA’s known error rates, to infer on a balance of probabilities that the AI's malfunction caused the injury. Analogising to W v ACC, the patient can sidestep the problem of a multiplicity of possible causes. Instead of proving that the developer's data or the hospital's monitoring was the sole or direct cause of the misdiagnosis, causation is satisfied if the use of the AI system "materially contributed" to the injury. Without the novus actus interveniens doctrine, the ACC cannot claim that the inherent unpredictability of the algorithm broke the chain of causation. In contrast, the many hands problem and concept of novus actus interveniens bar claimants in Singapore from holding hospitals and developers liable for the harm caused by SERA. Hence, the above principles of the New Zealand scheme are better than the Singapore tort system in delivering more timely and available compensation. However, the ACA is highly restrictive in black-box AI causation cases due to the exclusion of injuries that are wholly or substantially caused by the patient’s underlying health condition. 109 One downside is that operating on a fixed state budget, the ACC frequently relies on the ambiguity of this provision to deny claims in borderline cases and thereby reduce its expenditure. 110 In Robertson v ACC (“Robertson”), the patient suffered from IgA nephropathy, a chronic kidney disease.111 After a two-year delay in diagnosis, she had developed end-stage renal failure by the time she was referred to a specialist. Although the progression would have been delayed if she had received timely diagnosis and treatment, the court held that the end-stage renal failure was the "inevitable consequence" of the patient's underlying IgA nephropathy. 112 As the delay in diagnosis

108

Ambros (n 97) [89]-[92], [105]. ACA, s 32(2)(a). 110 Andrew (n 82) 36. 111 [2011] NZACC 327 [2]. 112 Robertson (n 111) [25]–[26]. 109

184

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

merely amounted to an acceleration of the inevitable, the injury was wholly or substantially caused by the underlying condition and the ACC denied compensation. 113 Hence, if SERA's missed prediction merely accelerated the inevitable, the ACC would likely treat the injury as wholly or substantially caused by the underlying condition. This leaves the patient uncompensated by the state. On the other hand, Professor Todd defends the ACC’s application of the underlying health condition provision. Although excluding general illnesses from a no-fault scheme inevitably creates borderline anomalies, this divide remains a practical necessity to ensure victims of accidents receive adequate compensation without the hurdles of tort litigation. 114 Expanding the framework to cover all diseases would financially collapse the scheme, effectively reducing it to a generalised social security system and forcing a substantial reduction in individual payouts. 115 Professor Manning argues that despite the 2005 no-fault reforms, "notions of negligence have crept back into the interpretation" of the ACC treatment injury criteria. 116 She specifically criticises how this occurs within the ordinary consequence exclusion.117 To determine if an injury is an ordinary consequence, decision-makers inherently evaluate the treatment that should have been given, effectively reintroducing a fault-based benchmark.118 The Court of Appeal’s decision in Adlam v ACC illustrates this dynamic. A child suffered cerebral palsy because an emergency caesarean was not performed earlier. 119 The Court held there is no failure to treat at all unless an alternative "could and should have been given having regard to the clinical indications at the time". 120 Since withholding intervention accorded with clinical knowledge at the time, the ensuing harm was an ordinary consequence of the non-treatment.121 As Professor Manning warns, making ACC cover contingent on whether a doctor departed from an objective clinical standard forces claimants to prove fault, undermining the scheme's non113

Robertson (n 111) [28]. Todd (n 87) 1211. 115 Todd (n 87) 1211. 116 Joanna M Manning, ‘Plus ça change, plus c'est la même chose: Negligence and Treatment Injury in New Zealand's Accident Compensation Scheme’ (2014) 14(1–2) Med Law Intl 22, 24. 117 ACA, s 32(1)(c); Manning (n 116) 51. 118 Manning (n 116) 51. 119 [2017] NZCA 457 [4(a)]. 120 Adlam (n 119) [61]. 121 Adlam (n 119) [73], [76]. 114

185

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

adversarial purpose.122 While the scheme abolishes the inquiry into who was to blame, it has not abolished the inquiry into what should have been done. Such an approach would reintroduce fault into the scheme, along with the evidential difficulties of proving fault, ultimately undermining patient compensation. C.

Deterrence and Patient Safety

Clinicians In New Zealand, rather than relying on the threat of civil litigation to deter unsafe practices, the scheme promotes patient safety through a "just culture" of open disclosure and learning. 123 This framework contrasts with Singapore’s approach, which relies on medical negligence law to deter automation bias. The Health and Disability Commissioner and the Medical Council serve as the primary mechanisms for accountability. 124 By removing the threat of compensatory damages, the ACA scheme fosters an environment where clinicians are more likely to honestly disclose when SERA misdiagnoses a patient. 125 This enables the health sector to trace and learn from the AI's failures rather than hiding them to avoid lawsuits.126 Simultaneously, clinicians face deterrence through Health Practitioners Disciplinary Tribunal disciplinary proceedings if they blindly trust an AI without exercising appropriate clinical judgment.127 Hence, professional standards are maintained through both a culture of learning and also the threat of disciplinary proceedings. Singapore likewise maintains professional discipline under the Medical Registration Act 1997. 128 The difference is that Singapore’s tort of negligence provides a deterrent that is linked to

122

Manning (n 116) 25. Marie Bismark, ‘No-Fault Compensation for Treatment Injury in New Zealand’ (Unpublished manuscript, Tel Aviv University 12 March 2008), 2. 124 Katharine Wallis, ‘New Zealand's 2005 'No-Fault' Compensation Reforms and Medical Professional Accountability for Harm’ (2013) 126(1371) NZMJ 33, 33. 125 Bismark (n 123) 9. 126 Bismark (n 123) 1. 127 Stephen Todd, ‘Medical Injury in New Zealand’ in Dobrochna Bach-Golecka (ed), Compensation Schemes for Damages Caused by Healthcare and Alternatives to Court Proceedings: Comparative Law Perspectives (Springer 2021) 327. 128 Medical Registration Act 1997 (2020 Rev Ed). 123

186

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

compensation. Both New Zealand and Singapore fulfil the same goal of deterring clinicians from automation bias through different legal methods. Hospitals Patient safety requires hospitals to implement procedural validation and continuous monitoring to safely use AI. Under section 284 of the ACA, if ACC claims data reveals a risk of public harm from a hospital’s AI system, the ACC must report this to safety authorities to trigger systemic interventions.129 Hospitals are thus incentivised to invest in the continuous validation and other procedures required to safely monitor opaque black-box systems such as SERA. Singapore has no equivalent bridge from compensation to regulation. The product is licensed before deployment under the Health Products Act 2007 and the hospital is licensed under the Healthcare Services Act 2020. 130 Deployment standards sit in non-binding guidance under the AIHGle and a settled private civil claim triggers no automatic regulatory risk report. The New Zealand deterrence function regarding hospitals is more effective than Singapore’s, whereby the difficulty of establishing the chain of causation allows the hospital to avoid legal liability and inhibits the deterrence effect. Developers In New Zealand, the ACA’s statutory bar pursuant to section 317 prevents patients from suing for compensatory damages. 131 Developers are thus shielded from traditional product liability claims. Under the Supreme Court's ruling in Couch v Attorney-General (“Couch”), the statutory bar does not prevent claims for exemplary damages. 132 Hence, a patient can seek exemplary damages against an AI developer to punish and deter outrageous conduct. However, Couch strictly confines exemplary damages to torts committed intentionally or with "subjective recklessness".133 To hold a developer liable, a claimant must prove the developer

129

Todd (n 87) 1205. Health Products Act 2007 (Singapore) ss 15(1), 30; Healthcare Services Act 2020 (Singapore) ss 8(1), 11(1). 131 ACA, s 317. 132 [2010] NZSC 27, [2010] 3 NZLR 149 [89]. 133 Couch (n 132) [259]. 130

187

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

"deliberately and outrageously ran a consciously appreciated risk of causing personal injury". 134 Proving that a developer subjectively foresaw and outrageously ignored specific training risks is an almost insurmountable evidentiary hurdle due to the unpredictability of the AI system. Consequently, the high Couch threshold provides an extremely weak deterrent against developers deploying poorly trained medical AI. This result of low deterrence is similar to Singapore and threatens patient safety in the long run. VI.

A MODIFIED NO-FAULT TRANSPLANT

A functionalist comparison of Singapore's tort system and New Zealand’s ACA reveals a shared vulnerability. Both impose substantial transaction costs on patients. Singapore does so expressly through proof of fault and causation and New Zealand does so residually, through fault reasoning that creeps into the ordinary consequence exclusion. Both regimes deter clinicians from automation bias, yet both fail to deter developers from neglecting procedural safeguards and leave hospital compliance to government oversight, which New Zealand at least reinforces through section 284 reporting. A modified transplant to Singapore should retain what each system does well and correct what each does poorly. The case for a no-fault transplant to Singapore rests on transaction costs and patient welfare. Administration consumes less than 10% of ACC expenditure against 50 to 60% in malpractice systems, straightforward claims resolve in weeks rather than years, and a larger share of injured patients receives compensation. 135 These advantages are exemplified in the black-box scenario. The SERA patient cannot untangle whether poor training data, deficient monitoring, or the model's own unforeseeable failure produced the missed prediction, so litigation spending is wasted on attempting proof amid inference opacity. A no-fault fund redirects that expenditure into timely compensation.136

134

Couch (n 132) [179]. Marie Bismark and Ron Paterson, ‘No-Fault Compensation in New Zealand: Harmonizing Injury Compensation, Provider Accountability, and Patient Safety’ (2006) 25(1) Health Aff 278, 279 and 281. 136 Bismark and Paterson (n 135) 281. 135

188

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

The transplant proceeds through judicious ecumenism, borrowing foreign best practice while preserving domestic equilibrium. 137 The triage asks whether the clinician had, and discharged, a meaningful HITL role. Where meaningful oversight was possible and the clinician failed to exercise it, negligence governs. This preserves the accountability that anchors the trust and confidence of the doctor-patient relationship.138 Where clinicians have fulfilled their HITL role, or opacity left no meaningful role to discharge, the clinician is shielded and the claim enters the fund. The fund still pays the patient regardless of fault, but recoups its costs from developers and deployers through graduated contributions, because control over the risk of harm resides with them. The developer selects the training data and the update pipeline. The deployer validates and monitors the system in its clinical environment. The clinician, as the Deutz analysis established above, cannot conduct an intermediate examination capable of revealing a latent defect. Concentrating responsibility on the parties able to prevent the error places deterrence where it can operate and stops clinicians from becoming liability sponges.139 Where no avoidable noncompliance can be identified, the claim is simply absorbed by the general pool. A no-fault scheme still requires its own deterrence function, because compensation meets patients' financial needs only and accountability must be secured elsewhere. 140 The scheme builds deterrence into its funding. Developers and deployers finance the fund through contributions that are reviewed annually and graduated according to each contributor's adherence to the AIHGle, assessed through regular audits.141 A developer with weak dataset governance or a hospital without post-deployment monitoring pays a loaded contribution and sustained compliance earns a smaller contribution. This grading converts the AIHGle from soft guidance into a pre-emptive market regulation mechanism, priced on the risk of patient harm before the harms materialise. The transplant must also modify the underlying condition exclusion. Applied as in Robertson, section 32(2)(a) denies cover to the patients who present with a deteriorating condition. Todd’s caution that the fund cannot underwrite disease itself is sound. Instead of a complete bar, the fund

137

Basil C Bitas, ‘Comparative Theory, Judges and Legal Transplants: A Practical Lesson from Singapore and its Relevance to Transnational Convergence’ (2014) 26 SACLJ 50, 60. 138 Stanley Yeo, “The Standard of Care in Medical Negligence Cases” (1983) 25 Mal L Rev 30, 48. 139 Crootof, Kaminski and Price (n 51) 483. 140 Bismark and Paterson (n 135) 278. 141 Ministry of Health, Health Sciences Authority and Synapxe (n 12) 5 and 17.

189

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

should apportion. Assessors can fix the probable course of the underlying condition under proper care, value the difference between that course and the outcome suffered, and compensate for that increment. Assessment under the scheme must not reintroduce fault. As shown above, negligence reasoning has crept back into treatment injury adjudication. Manning’s recommendation should be adopted, whereby avoidability is judged in hindsight. The assessor should ask, based on the information available at the date of the claim and with knowledge of the treatment outcome, whether different performance or an available alternative would have avoided the injury. 142 This approach contrasts to the inquiry into what the reasonable practitioner judged at the time. Hence, the inquiry seeks to compare outcomes instead of fault and eliminates the hefty transaction costs associated with proving fault in the first place. Finally, the scheme regulates a temporary state of technology and should retire when technology advances eliminate the opaque nature of black-box AI. Explainable AI (“XAI”) is advancing, but a validation gap persists. Current explanation methods persuade clinicians with plausible outputs, but they fail to provide consistent, verifiable proof of how the AI actually reasoned. 143 A sunset clause should mandate review at fixed intervals on the advancements of XAI. Once explanations reliably trace an erroneous output to its origin, inference opacity no longer hinders proof of fault and causation. Thus, the evidentiary premise of the scheme lapses, and fault-based recovery can resume without the prohibitive transaction costs to patients. VII.

CONCLUSION

Black-box AI does not defeat the tort system everywhere, but only where inference opacity removes the patient’s ability to prove fault and causation. Singapore's negligence regime compensates and deters when the clinician can check the machine. When the clinician cannot, the patient likely goes uncompensated, while the developer and deployer remain undeterred. New Zealand supplies the architecture of a no-fault approach together with case studies on implementation pitfalls, as it excludes high-risk claimants and fault reasoning creeps into the

142

Manning (n 116) 37–38. Kai Zhang and others, 'A Comprehensive Review of Explainable Artificial Intelligence in Healthcare: Methods, Evaluation, and Clinical Integration' (2026) 29 iScience 115026, 24. 143

190

Singapore Comparative Law Review

Statutory Sutures For The Liability Gap

system. The modified transplant takes the approach and corrects the execution. Triage preserves clinician accountability, graduated contributions price safety into the market, retroactive avoidability keeps fault out of adjudication, apportioned cover protects high-risk patients, and a sunset clause returns the field to tort once black-box AI is no longer opaque.

191

Join a Truly Global Firm

Key Dates

Our global management committees unite our offices around the world to drive the firm forward. Wherever you are in the world, our vast global coverage will mean you have access to a borderless network of expertise.

2027 SG Vacation Scheme

Applications close: February 1, 2027

Benefit From Extensive Resources and Training

To find out more about our Summer Vacation Scheme and Training Contract, please visit LWCAREERS.COM

From our signature academies focused on skills development, to our CareerDesign program helping you shape the direction of your career, Latham invests in your future.

Feel Empowered to Actively Contribute

Applications open: October 1, 2026

Singapore

Associates participate in leadership, including in our global committees. Reviews, bonuses, and promotions are decided by our Associates Committee, comprised equally of associates and partners.

Latham & Watkins operates worldwide as a limited liability partnership organized under the laws of the State of Delaware (USA) with affiliated limited liability partnerships conducting the practice in France, Hong Kong, Italy, Singapore, and the United Kingdom and as an affiliated partnership conducting the practice in Japan. Latham & Watkins operates in Israel through a limited liability company, in South Korea as a Foreign Legal Consultant Office, and in Saudi Arabia through a limited liability company.

Singapore Comparative Law Review

Unbroken Will

UNBROKEN WILL – ASSESSING THE FUTURE OF VOLUNTARINESS IN SINGAPORE’S CRIMINAL PROCEDURE CODE AND ADMISSIBILITY REGIME Nickolaus Ng* ABSTRACT This article analyses Singapore’s confession admissibility framework under Section 258 of the Criminal Procedure Code, arguing that the existing voluntariness test already sufficiently addresses reliability concerns without needing to adopt the United Kingdom's PACE approach. It concludes that a major overhaul of the Singaporean model is unnecessary, as the current system effectively balances procedural efficiency with protections against unfairness. I.

INTRODUCTION

Singapore’s confession admissibility regime, grounded in voluntariness under s 258 of the Criminal Procedure Code (“CPC”) and supplemented by a narrow oppression doctrine, prioritises certainty and investigative efficiency but remains normatively underdeveloped in its treatment of reliability and fairness. While judicial recognition of oppression has softened the rigidity of the voluntariness test, oppression in Singapore operates only as a subsidiary mechanism to negate free will, rather than as an independent safeguard of evidential integrity. By contrast, the United Kingdom’s reliability-centred framework under the Police and Criminal Evidence Act 1984 (“PACE”) reconceptualises admissibility as a systemic inquiry into trustworthiness and procedural legitimacy. This divergence reflects a fundamental normative choice between formal consent and substantive justice, raising the question of whether Singapore’s current approach remains adequate in light of modern understandings of interrogation dynamics and wrongful conviction risks. Nevertheless, the recalibration of Singapore’s voluntariness test toward a reliability-centred model is unnecessary. First, the voluntariness standard already indirectly safeguards reliability: a statement extracted through coercion, inducement, or oppression is inherently suspect, and exclusion under Section 258 CPC ensures that unreliable confessions do not enter the evidentiary record. Second, Singapore’s criminal process contains complementary safeguards, namely the judicial discretion under Section 258(3) CPC, and this collectively mitigates wrongful conviction risks without destabilising investigative certainty. Finally, importing a *

Advocate and Solicitor of the Supreme Court of Singapore

192

Singapore Comparative Law Review

Unbroken Will

reliability-based framework akin to PACE would introduce unnecessary complexity and uncertainty into Singapore’s streamlined evidentiary regime, potentially undermining investigative efficiency without demonstrable gains in fairness. Thus, the existing voluntariness test, supplemented by oppression, remains sufficient to balance evidential integrity with procedural economy, rendering wholesale modification unwarranted. II.

TENETS OF THE CURRENT ADMISSIBILITY REGIME IN SINGAPORE

The admissibility regime in Singapore, as it currently stands, primarily targets the principles of reliability, protection of accused persons and to serve as a disciplinary function against law enforcement agencies and the Prosecution. These are explored in turn. A. Reliability Principle The reliability principle reflects the concern that criminal trials should not be distorted by admitting evidence that is inherently untrustworthy. This rationale was first expressed in the UK in The King v Warickshall, where Eyre B remarked that confessions “…forced from the mind by the flattery of hope, or by the torture of fear, come in so questionable a shape when they are to be considered as evidence of guilt, that no credit ought to be given to them”.1 Only confessions made freely and voluntarily were considered to carry the highest evidential value. The underlying assumption is that a rational person would not voluntarily make a statement against their own interest unless it were true. In Mortimer v Mortimer, confessions were described as ranking “highest in the scale of evidence”. 2 Yet confessions may become unreliable when induced by hope or fear. As clarified in Regina v Baldry, the law does not presume such statements are always false, but judges exclude them because of the danger in admitting them.3 This caution is particularly important in criminal cases, where miscarriages of justice must be avoided. The risk of false confessions is well-documented, as seen in the wrongful convictions of the “Guildford Four” and “Birmingham Six” in the UK.4 In Singapore, similar concerns arose in PP v Somporn Chinphakdee [1994] SGHC 209, where the High Court acquitted the

The King v Warickshall (1783) 1 Leach 263. Mortimer v Mortimer [1920] 2 Hag Con 310. 3 Regina v Baldry (1852) 2 Den 430. 4 R v McIlkenny and Others [1991] 93 Cr App R 287. 2

193

Singapore Comparative Law Review

Unbroken Will

accused after finding his confession involuntary, 5 and in 1993 when a false confession led the Prosecution to withdraw charges against Samat Dupree. 6 The reliability principle remains the central justification for the exclusionary rule on confessions. Lord Griffiths in Lam Chi-Ming noted that the “fruit of the poisoned tree” doctrine developed from this focus on reliability. 7 Under Section 76(4) of the UK Police and Criminal Evidence Act 1984, a confession may be excluded for involuntariness, but facts discovered because of this may still be admissible. This reflects the assumption that a confession should not be wholly disregarded if corroborated by independent facts. The Criminal Law Revision Committee acknowledged that only the reliability principle could justify this doctrine. Singapore has gone further by enacting Section 27 of the Evidence Act 1893 (“EA”), which allows parts of a confession to be admitted if they lead to the discovery of relevant facts. Still, reliability is not the sole rationale behind the voluntariness test. If reliability were the only consideration, courts would inquire into the truth of the confession to decide admissibility. Yet the Privy Council in Wong Kam-ming v The Queen held that it was improper for the prosecution to cross-examine on the voir dire to establish the truth of the confession.8 Lord Edmund-Davies, citing the Canadian case R v Hnedish, endorsed Hall CJ’s view that:9 I do not see how under the guise of ‘credibility’ the court can transmute what is initially an inquiry as to the ‘admissibility’ of the confession into an inquisition of an accused. That would be repugnant to our accepted standards and principles of justice; it would invite and encourage brutality in the handling of persons suspected of having committed offences.

The Privy Council has emphasised that the reliability of a confession should be assessed separately from its admissibility at the voir dire stage, with reliability left to the jury once admissibility is determined. 10 This shows that the exclusion of involuntary statements rests on more than just reliability. In Singapore, both Section 122(5) CPC and Section 24 EA are closely tied to broader concerns of justice and propriety. The High Court recognised this in Public Prosecutor v Sng Siew Ngoh, where Yong Pung How CJ explained that Section 122 CPC was designed to regulate police conduct:

Public Prosecutor v Somporn Chinphakdee [1994] SGHC 209 Samat Dupree case (1993) (unpublished). 7 Lam Chi-Ming v R [1991] 2 AC 212. 8 Wong Kam-ming v The Queen [1980] AC 247. 9 R v Hnedish (1960) 126 CCC 126. 10 Wong Kam-ming v The Queen [1980] AC 247. 6

194

Singapore Comparative Law Review

Unbroken Will

the basis of the proviso… is not rooted in the reliability or otherwise of the statement made to the police, but is intended clearly to prevent any impropriety on the part of the interrogators.11

Notably, not all unreliable confessions are excluded under the voluntariness test, which suggests that other rationales underpin the exclusionary rule. Mirfield has questioned the claim that reliability is “quite clearly” the governing rationale, pointing out that a confession obtained through strong spiritual inducement may be unreliable, yet the voluntariness test would not apply since only inducements of a temporal nature are relevant. 12 Likewise, interrogation methods such as the “Inbau and Reid tactics” involve stage-managing the suspect’s environment, through room design, selective disclosure of evidence, or sympathetic framing to elicit confessions.13 Although no overt coercion is used, such techniques may still produce false confessions, yet the voluntariness test does not capture them. Mirfield further argues that the interrogation process itself increases the likelihood of confessions regardless of their truth. 14 If reliability were the sole concern, the exclusionary rule would exclude all confessions whose truth was doubtful. Instead, the common law voluntariness test has consistently focused on threats, promises, or inducements, and their effect on the accused’s mind. This demonstrates that fairness, propriety, and the regulation of police conduct are also central to the voluntariness test. The key issue for Singapore is whether the law should continue upholding these broader rationales alongside reliability. B. Protective Principle Ashworth first introduced the protective principle in 1977, arguing that suspects possess rights which the legal system must respect, and that courts should exclude evidence obtained through improper means.15 A rights infringement therefore provides a prima facie justification for exclusion. Modern English cases have linked this principle to the privilege against selfincrimination, emphasising the need to protect individuals from pressure by officials to incriminate themselves. As Lord Diplock explained in R v Sang:

Public Prosecutor v Sng Siew Ngoh [1996] 2 SLR(R) 38. Peter Mirfield, Silence, Confessions and Improperly Obtained Evidence (Clarendon Press 1997). 13 ibid. 14 ibid. 15 Andrew Ashworth, 'Excluding Evidence as Protecting Rights' (1977) Crim LR 723. 12

195

Singapore Comparative Law Review

Unbroken Will

The underlying rationale of this branch of the criminal law, though it may originally have been based upon ensuring the reliability of confessions is, in my view, now to be found in the maxim nemo debet prodere se ipsum… or in its popular English mistranslation ‘the right to silence’.16

It can be questioned whether the voluntariness test should protect this privilege. The theoretical difficulty of justifying a suspect’s right not to answer police questions, and the practical challenge of determining when pressure breaches the privilege. Strict enforcement would effectively make police interrogation pointless. Nevertheless, the Court of Appeal in PP v Mazlan bin Maidun held that the privilege against self-incrimination was not a fundamental rule of natural justice, and its breach did not affect admissibility.17 Yet the protective principle extends beyond this privilege. As Mirfield observed, it is concerned with maintaining minimum standards in the treatment of suspects, encompassing broader humanitarian rights. 18 Ashworth similarly explained that the exclusionary rule reflects the courts’ recognition that suspects should not be subject to inducement or oppression.19 Thus, the privilege against self-incrimination is only one aspect of the wider protective principle. Even if diminished in Singapore, the broader rationale remains valid. The voluntariness test should continue to uphold these protective concerns alongside reliability. In other judicial decisions, courts have increasingly framed the protective principle in terms of fairness to the accused. Lord Hailsham’s remarks in Wong Kam-ming v The Queen are often cited:20 Any civilised system of criminal jurisprudence must accord to the judiciary some means of excluding confessions or admissions obtained by improper methods. This is not only because of the potential unreliability of such statements, but also, and perhaps mainly, because in a civilised society it is vital that persons in custody or charged with offences should not be subjected to ill treatment or improper pressure in order to extract confessions.21

Similarly, Lord Salmon in R v Sang grounded the exclusion of involuntary confessions on fairness: “A confession by an accused which has been obtained by threats or promises is

R v Sang [1980] AC 402. Public Prosecutor v Mazlan bin Maidun [1992] 1 SLR(R) 772. 18 Peter Mirfield, Silence, Confessions and Improperly Obtained Evidence (Clarendon Press 1997). 19 Andrew Ashworth, 'Excluding Evidence as Protecting Rights' (1977) Crim LR 723. 20 Wong Kam-ming v The Queen [1980] AC 247. 21 Wong Kam-ming v The Queen [1980] AC 247. 17

196

Singapore Comparative Law Review

Unbroken Will

inadmissible as evidence against him, because to admit it would be unfair”.22 This reasoning, echoed in Commissioners of Customs and Excise v Harz, underscores that the protective principle safeguards a suspect’s right not to be subjected to coercion or ill-treatment.23 The weakening of the privilege against self-incrimination does not, therefore, diminish the continuing relevance of the protective rationale. It is also arguable that the protective and reliability principles can both be satisfied through the voluntariness test. While an involuntary confession may sometimes be true, as Michael Hor observed: if a suspect says something when he may have chosen not to do so, this raises suspicions as to the voluntariness of the suspect’s confession.24 This highlights the apparent tension between reliability and protection since protecting the accused may mean excluding confessions that may be reliable. Yet in practice, this conflict is more illusory than real. At the voir dire stage, the court cannot know whether a confession is true or false; it can only infer potential unreliability from the presence of threats or inducements. Thus, excluding a confession on protective grounds does not necessarily sacrifice reliability. A confession excluded under the voluntariness test may be reliable, but this can never be known with certainty. At such a preliminary stage, caution is essential. C. Disciplinary Principle The disciplinary principle focuses on preventing improper police practices. Excluding confessions obtained through misconduct is intended to discourage investigators from resorting to such methods. When Stephen drafted the Indian Evidence Act 1872,25 his prohibition on police statements was motivated by concerns that torture would otherwise persist. Singapore’s original CPC and the EA adopted this approach, reflecting the same rationale. Today, however, the disciplinary principle appears only in diluted form within the voluntariness test under Section 122(5) CPC. Unlike earlier provisions, the current law does not assume that all policeextracted statements involve misconduct. Instead, it condemns only threats, inducements, or promises serious enough to undermine free will.

R v Sang [1980] AC 402. Commissioners of Customs and Excise v Harz [1967] 1 AC 760. 24 Michael Hor, “The Confession Rule in Singapore” (1995) 7 SAcLJ 240 at 249. 25 Act No.1 of 1872. 23

197

Singapore Comparative Law Review

Unbroken Will

Unlike the reliability and protective principles, which can coexist under the voluntariness test, the disciplinary rationale is harder to reconcile. Reliability and protection focus on the effect of inducements on the accused, while discipline targets police impropriety itself. For instance, interrogation tactics may not amount to threats or promises sufficient to render a confession involuntary. On reliability grounds, the statement would be admissible; on protective grounds, the accused’s rights would not be infringed. Yet under the disciplinary principle, exclusion may still be justified to signal disapproval of improper practices. Strict adherence to discipline, however, risks excluding confessions that are otherwise reliable. It is posited that reliability is compromised when courts attempt to redress impropriety, and there is little consensus on the scope of misconduct that should trigger exclusion. Moreover, the deterrent effect is doubtful. Section 27 EA allows facts discovered through inadmissible confessions to be admitted, diluting any deterrent impact. As Mirfield observed, deterrence is only effective if supported by internal police procedures to reinforce judicial signals. 26 Accordingly, the exclusionary rule should not rest primarily on the disciplinary rationale. Its relevance is better understood in conjunction with the protective principle: statements should be excluded to shield the accused from improper treatment, not simply to deter police misconduct. This does not deny the disciplinary rationale’s presence in the voluntariness test. Public Prosecutor v Sng Siew Ngoh illustrates judicial concern about improper threats or promises, but impropriety should not be the chief basis for exclusion. 27 Instead, mandatory exclusion should be grounded in protection and reliability, while improprieties such as entrapment or deception are more appropriately addressed through judicial discretion. It is observed that all three tenets of reliability, protective, and disciplinary functions form the triangular base for the admissibility regime in Singapore and form the basis for the voluntariness test in Singapore. While reliability is not expressly articulated as an independent admissibility criterion, the protective function against coercion and the disciplinary function deterring improper police conduct converge to indirectly secure reliability. In practice, the voluntariness test already embodies these three dimensions: it excludes confessions tainted by inducement or oppression (i.e. protective), signals to law enforcement that improper tactics will nullify evidentiary gains (i.e. disciplinary) and ensures that only statements freely made are admitted (i.e. reliability). Thus, the triangular foundation of the exclusionary rule is not absent 26 27

Peter Mirfield, Silence, Confessions and Improperly Obtained Evidence (Clarendon Press 1997). Public Prosecutor v Sng Siew Ngoh [1996] 1 SLR 143.

198

Singapore Comparative Law Review

Unbroken Will

in Singapore but embedded within the voluntariness framework itself, obviating the need for doctrinal modification. However, no system is without its flaws, and this article now moves on to explore the difficulties that are inherent in the voluntariness test. D. Difficulties Inherent in the Voluntariness Test Section 258(3) CPC is substantially identical to the former Section 24 EA and its predecessor in the Evidence Ordinance 3 of 1893. However, the terminology of Section 258(3) CPC is vague, and its fragmented drafting makes the test overly technical and difficult to apply, thereby undermining reliability. 28 For instance, the Singapore Court of Appeal (“SGCA”) expressed dissatisfaction with the wording of the former Section 24 EA (now replicated in Section 258(3) CPC), holding that the phrase “having reference to the charge” must be purposively construed to provide logical coherence and avoid injustice. 29 Terms such as “having reference to the charge,” “person in authority

”, “inducement, threat or promise

temporal nature

” and “any advantage or avoid any evil of a

” together with the implied objective and subjective elements in Section

258(3) CPC, have all been judicially considered. 30 The weakness of the voluntariness test has also been highlighted by the SGCA’s willingness to exclude statements that satisfy the voluntariness requirement but are nonetheless unreliable. 31 Beyond its terminology, the voluntariness test is too narrow to capture other forms of unreliable evidence. Consider the following scenarios: A and B are charged with theft, but B is innocent. If A promises B the proceeds of the theft upon release, B’s confession would be voluntary and admissible under Section 258(3) CPC since A is not a person in authority though clearly unreliable. Courts have held that self-perceived inducement does not invalidate a confession.32 In DPP v Ping Lin, ruling that a "self-generated" hope or impression cannot satisfy the voluntariness test if no person in authority actually said or did anything to excite that expectation.33 Nevertheless, such circumstances still raise reliability concerns.

Muhammad bin Kadar v Public Prosecutor [2011] 3 SLR 1205. ibid. 30 ibid. 31 ibid. 32 Lu Lai Heng v Public Prosecutor [1994] 1 SLR(R) 1037 33 DPP v Ping Lin [1976] AC 574; [1975] 3 All ER 175 at 189. 29

199

Singapore Comparative Law Review

Unbroken Will

Apart from reliability, other principles have been recognised in common law jurisdictions. One such principle is the “judicial integrity principle

” which prevents courts from relying on

evidence obtained through serious impropriety. The judicial integrity principle was considered by VK Rajah J Wong Keng Leong Rayney v Law Society of Singapore.34 Unlike reliability, these principles are not primarily concerned with the correctness of judicial outcomes. While they may overlap with voluntariness, they are not foundational to the law of evidence, which focuses on reliability. For instance, Section 258(3) CPC, Explanation 2(a), validates confessions obtained through deception or promises of secrecy. In How Poh Sun v PP, the SGCA held that police entrapment is not a basis for excluding evidence. 35 In Law Society of Singapore v Tan Guat Neo Phyllis, the High Court emphasised that the EA is unconcerned with the manner in which evidence is secured. 36 The voluntariness test has been abrogated in many common law jurisdictions. 37 The main criticism of the voluntariness test is that it fails to uphold the protective principle, as it is inherently inconsistent.38 It seems illogical to exclude confessions on the basis of voluntariness, since in reality very few confessions are genuinely made out of free will. 39 Recognising this, Lord Lane CJ observed in R v Rennie: Very few confessions are inspired solely by remorse. Often the motives of an accused person are mixed and include a hope that an early admission might lead to an earlier release or a lighter sentence. If it were the law that the mere presence of such a motive, even if prompted by something said or done by a person in authority, led inexorably to the exclusion of a confession, nearly every confession would be rendered inadmissible.40

The UK Criminal Law Revision Committee did not favour the common law test, noting that “any threat or inducement, however mild or slight, uttered or held out by a person in authority makes a resulting confession inadmissible”.41 In Singapore, it has been observed that “there really is no free will and it is a rare suspect indeed who would prefer to confess,” and that “the courts are invited to search for the free will which no longer exists at the expense of focusing

Wong Keng Leong Rayney v Law Society of Singapore [2007] SGCA 42. How Poh Sun v Public Prosecutor [1991] 1 SLR 220. 36 Law Society of Singapore v Tan Guat Neo Phyllis [2007] SGHC 207. 37 For example, the United Kingdom, Australia and New Zealand. 38 Andrew Ashworth, The Criminal Process: An Evaluative Study, 2nd ed., 1998, at p. 244. 39 Paul Roberts & Adrian Zuckerman, Criminal Evidence, 2nd ed., 2010, at p. 446. 40 R v Rennie [1982] 1 WLR 64 at 69. 41 Criminal Law Revision Committee, Eleventh Report: Evidence (General), Cmnd 4991, 1972, at para. 112. 35

200

Singapore Comparative Law Review

Unbroken Will

on reliability”.42 Hor also argued that the voluntariness test, being a translation of the flawed privilege against self-incrimination, fails to address reliability concerns.43 He later examined the concept of oppression, pointing out its conceptual difficulties within the confession regime and noting that it remains tied to the futility of the protection against self-incrimination.44 Oppression was described in R v Prager as “something which tends to sap, and has sapped, that free will which must exist before a confession is voluntary

”45 and oppressive questioning

as “questioning which… so affects the mind of the subject that his will crumbles and he speaks when otherwise he would have stayed silent.”46 In this context, Hor concludes that there is, in reality, no free will when a suspect confesses.47 In Singapore, however, courts have interpreted oppression through the lens of the objective limb of the voluntariness test. For example, in Fung Yuk Shing v PP, the SGCA considered whether the alleged oppressive conduct amounted to a threat or inducement of such seriousness that any resulting statement would automatically be involuntary. 48 Through this line of reasoning, the SGCA effectively subjects oppression to an objective test of reasonableness. Thus, Singapore courts do not interpret “voluntariness” literally. The word itself does not appear in Section 122(5) CPC, it was borrowed from the common law exclusionary rule. As a result, there is relatively less uncertainty in Singapore about the scope and elements of the voluntariness test. III.

THE POSITION IN THE UNITED KINGDOM

The exclusionary rule in the UK began as a common law doctrine, its clearest early statement appearing in Warickshall in the mid-18th century. That case articulated the “reliability principle ” holding that a confession obtained through hope or fear was so questionable that it should not be credited. Yet, as Mirfield later noted, judges in subsequent cases avoided speculating on whether a confession was actually false. 49

Michael Hor, “The Confession Rule in Singapore” (1995) 7 SAcLJ 240, at 249. ibid., at 250. 44 ibid., at 252. 45 R v Prager [1972] 1 WLR 260, at 266. 46 ibid. 47 Michael Hor, “The Confession Rule in Singapore” (1995) 7 SAcLJ 240, at 253. 48 Fung Yuk Shing v Public Prosecutor [1993] 2 SLR(R) 771. 49 Peter Mirfield, Silence, Confessions and Improperly Obtained Evidence (Clarendon Press 1997). 43

201

Singapore Comparative Law Review

Unbroken Will

Examples illustrate this development. In Cass’ Case,50 a publican promised favour to a suspect if he revealed stolen property, and Gould J ordered an acquittal, stating that even the slightest hope of mercy invalidated a confession. In Sexton’s Case,51 a suspect offered to confess in exchange for gin, and again the confession was excluded. This reflected a presumption that any inducement vitiated a confession regardless of reliability. The rule was eventually formalised in principle (e) of the 1964 Judges’ Rules, 52 which required that any statement be voluntary and not obtained through fear of prejudice, hope of advantage, or oppression by a person in authority. Dissatisfaction with this state of law led the UK Criminal Law Revision Committee in its Eleventh Report on Evidence to question the voluntariness test. 53 It highlighted cases where even mild inducements had led to exclusion and recommended modification: confessions should be excluded only if threats or inducements were likely, in the circumstances, to render them unreliable. Oppression, however, was to remain a separate ground of inadmissibility. Although Parliament did not immediately adopt these proposals, they were later reflected in Section 76(2) of the Police and Criminal Evidence Act 1984 (“PACE”). PACE effectively endorsed the Committee’s view that the exclusionary rule should rest on reliability, while maintaining oppression as an independent ground. Section 76 formed part of a broader framework: Section 78(1) gave courts discretion to exclude evidence if its admission would adversely affect fairness, and Section 77(1) specifically addressed the risk of unreliability in confessions by mentally impaired accused persons. Together, these provisions marked a shift towards reliability as the central principle, while preserving safeguards against oppression. IV.

IS IT PRACTICAL TO APPLY THE UK’S REFORMS TO SINGAPORE?

The legal backdrop in England stands in stark contrast to Singapore’s present framework governing confessions. In the UK, the Criminal Law Revision Committee was dissatisfied with the common law voluntariness test, which it regarded as unduly favourable to defendants. This leniency stemmed largely from the wording of the rule, which lacked an objective dimension. English courts were not required to assess the adequacy of a threat, inducement, or promise,

Cass’s Case (1784) 1 Leach 293. Sexton’s Case (1823) cited in George Chetwynd, Supplement to Burn's Justice of the Peace and Parish Officer (A. Strahan 1823) 103. 52 Practice Note (Judges' Rules) [1964] 1 W.L.R. 152. 53 Criminal Law Revision Committee, Eleventh Report: Evidence (General), Cmnd 4991, 1972. 51

202

Singapore Comparative Law Review

Unbroken Will

nor to determine whether the accused had reasonable grounds to believe that confessing would yield a benefit or avert harm. It was sufficient that the confession arose from fear of prejudice or hope of advantage. Singapore’s approach under Section 122(5) of the CPC does not suffer from this shortcoming. The voluntariness test, as articulated in Chai Chien Wei Kelvin v Public Prosecutor, clearly incorporates both subjective and objective components.54 Unlike the largely subjective English common law rules, Singapore's emphatically objective exclusionary principle was drawn from Stephen's Indian Evidence Act. Courts here have consistently declined to exclude confessions based on minor inducements, such as vague assurances of help or routine exhortations to tell the truth. If criticism is warranted, it is more likely to be that Singaporean courts have leaned towards the objective element, thereby making it exceedingly difficult for accused persons to demonstrate involuntariness. The English position prior to the PACE of 1984 is therefore far removed from Singapore’s jurisprudence. The rationale that prompted the UK to overhaul its voluntariness test does not apply in Singapore, where the statutory regime already imposes a stricter standard. Introducing an even more rigid framework would serve little purpose. The only persuasive argument for importing the UK model, despite Singapore’s distinct context, would be to give clearer effect to the reliability rationale. Yet even on this basis, legislative reform is not warranted. Shaping the exclusionary rule along the lines of Section 76 PACE does not necessarily advance the reliability principle articulated in Warwickshall. As the Criminal Law Revision Committee explained, the judge under the proposed reliability test was to imagine himself present during the interrogation, hearing the threat or inducement. He was then to weigh the strength of the inducement and the seriousness of the offence before asking: ‘… at the point when the threat was uttered or the inducement offered, any confession which the accused might make as a result of it would be likely to be unreliable”.55 This was reinforced by Section 76(1) PACE, which stipulates that once the test is satisfied, the confession must be excluded “notwithstanding that it may be true”. If Singapore were to adopt the PACE framework, the shift would be from examining the voluntariness of the accused’s confession to considering the probable effect of police conduct

54 55

Chai Chien Wei Kelvin v Public Prosecutor [1998] 3 SLR(R) 619. Criminal Law Revision Committee, Eleventh Report: Evidence (General), Cmnd 4991, 1972.

203

Singapore Comparative Law Review

Unbroken Will

on the truthfulness of the statement. While this places emphasis on reliability, it is doubtful that the test would operate very differently in practice. Both approaches are objective, requiring the court to impose its own assessment of the likely consequences of interrogation. The contrast between the UK common law rule and PACE is more significant, since under the former, English courts often stopped at identifying the existence of a threat or inducement without analysing its impact on the accused. By contrast, the difference between PACE and Singapore’s voluntariness test is relatively modest, making it likely that local courts applying the reliability test would continue to rely on voluntariness considerations. It is, after all, a logical inference that a confession extracted through coercion or inducement is also likely to be unreliable as per Section 122(5) CPC. This expectation is supported by English case law interpreting Section 76(2) PACE. In R v Barry,56 the Court of Appeal of England and Wales set out a structured approach. First, the judge must identify the relevant act or statement, considering all that was said and done by the police. Second, he must determine whether that conduct was likely to render unreliable any confession made in consequence, applying an objective test. Finally, he must ask whether the Prosecution has proved beyond reasonable doubt that the confession was not obtained as a result of the conduct, a factual inquiry to be approached with common sense. 57 The framework under PACE closely parallels the law as it currently stands in Singapore. In Public Prosecutor v Tan Boon Tat, the Court of Appeal held that courts must first determine whether any threat, inducement, or promise had prompted the confession. 58 This step is comparable to the initial stage in Barry, which requires identifying external factors capable of influencing the accused’s mind. The next inquiry in Singapore is whether the accused had reasonable grounds to believe that making the statement would help him avoid harm or secure an advantage.59 This is an objective assessment, much like Barry, though the standard applied is not explicitly tied to reliability. This fact-driven approach requires attention to the accused’s personal characteristics, the realities of police practice, the seriousness of the offence, and even the plausibility of the

R v Barry (1992) 95 Cr App R 384. ibid. 58 Public Prosecutor v Tan Boon Tat [1992] 1 SLR(R) 698. 59 ibid. 57

204

Singapore Comparative Law Review

Unbroken Will

confession itself. For example, in PP v Selvakumar Pillai s/o Suppiah Pillai,60 the District Judge found it implausible that the accused, charged with theft, would suddenly confess after refusing to do so in two earlier interviews. The suspicious circumstances of three officers leaving the accused alone with another officer before the confession, led to the inadmissibility of the confession being upheld by the High Court.61 Likewise, in Ng Ah Soi v PP,62 the High Court ruled that the statement was involuntary because of the absurdity of its content. The burden of proof also mirrors Barry: once voluntariness is challenged, the prosecution must prove beyond reasonable doubt that the confession was voluntary. In essence, the differences between PACE and Singapore’s CPC are minimal. PACE has also altered the outcomes of certain pre-PACE cases. For instance, in R v Richards,63 a policeman told a suspect it would be better to make a statement. It can be argued that the confession would likely be treated as reliable, although it may not have been done voluntarily by the defendant. Similarly, in R v Smith,64 a confession was made shortly after a regimental sergeant-major threatened to detain the company until someone admitted to murder. It is doubtful that an innocent person would have confessed reliably under such restrictive circumstances given the serious penalties for such an offence. Yet, under Singapore’s voluntariness test such threats or inducements would probably not be considered sufficiently serious to invalidate a confession. Accordingly, the reliability test does not significantly alter Singapore’s law on confessions. Given the vagueness of the reliability standard, it is highly likely that courts will continue to rely on the more detailed and structured framework in Section 122(5) CPC. Since the judicial evaluation of facts would remain largely unchanged, there is little justification for modifying the current law in Singapore along the lines of PACE. It may also be contended that the reliability test has value in preventing the wholesale exclusion of involuntary confessions that are, in fact, truthful. While this argument appears persuasive in theory, in practice the indeterminate nature of the reliability standard makes it doubtful that judges can identify such cases with any precision. By contrast, the voluntariness test provides

Public Prosecutor v Selvakumar Pillai s/o Suppiah Pillai [2004] SGDC 84. Public Prosecutor v Selvakumar Pillai s/o Suppiah Pillai [2004] 4 SLR 280. 62 Ng Ah Soi v Public Prosecutor [1996] 1 SLR 534. 63 R v Richards [1967] 51 Cr App R 266. 64 R v Smith [1959] 2 QB 35. 61

205

Singapore Comparative Law Review

Unbroken Will

clear and workable criteria, namely, threats or promises that create fear of harm or hope of advantage. It is submitted that this allows for an objective evaluation of the facts. Once the inquiry moves beyond voluntariness, however, it becomes exceedingly difficult to determine whether a statement made under coercion or inducement is nonetheless accurate. At the voir dire stage, reliability is virtually impossible to assess. As the UK Criminal Bar Association and the Release Lawyers’ Group observed in 1972, the reliability test requires judges to engage in speculation. 65 Apart from its doubtful utility, introducing the reliability test into Singapore could have problematic consequences. A judge faced with such a test might be tempted to conclude that a confession remains admissible even if made under threats or promises, so long as he believes the statement to be truthful and therefore reliable. This, however, misapplies the test, which requires consideration of the likely effect of the inducement on reliability, not an assessment of the confession’s actual truth. The danger is that a judge’s personal view of actual reliability may inadvertently influence the voir dire. In the UK, this conflation of potential and actual reliability has been treated as a serious error, since in cases where there is a jury, the jury alone should decide whether an admitted confession is reliable. The same issue should not be decided twice. Thus, in Wong Kam-ming v The Queen, the Privy Council held that cross-examination aimed at proving the truth of a confession was impermissible at the voir dire. As Lord Edmund-Davies explained: “If the defendant denies the truth of the confession… the question whether his denial is itself true or false cannot be ascertained until after the voir dire is over and the defendant’s guilt or innocence has been determined by the jury – an issue which the judge has no jurisdiction to decide.”

Although Singapore does not have juries, the risk of conflating admissibility with evidential weight remains. Reliability and admissibility are distinct inquiries that should not be blurred. As Yong CJ observed in PP v Huang Rong Tai, admissibility does not confer infallibility; a statement may still be rejected at trial for unreliability. 67 If reliability is pre-emptively

Criminal Bar Association and Release, Joint Memoranda on the Eleventh Report of the Criminal Law Revision Committee (1972); discussed in Ho Hock Lai, “The Concept of Voluntariness in the Law of Confessions” (1991) 3 SAcLJ 66 at 77. 66 Wong Kam-ming v The Queen [1980] AC 247 (Privy Council). 67 Public Prosecutor v Huang Rong Tai [2003] SGCA 1.

206

Singapore Comparative Law Review

Unbroken Will

addressed at the voir dire, the inquiry into admissibility risks becoming an indirect inquisition of the accused, which could prejudice the defence. UK case law demonstrates this confusion. In R v Cox, the court appeared to focus on actual reliability, with Lord Lane CJ providing the following clarification: 68 It seems to us that the true question was not whether the confession was unreliable or untrue so much as whether the confession, true or not, was obtained in consequence of anything done which was likely to render any confession unreliable, the burden being on the prosecution to prove beyond reasonable doubt that it was not so obtained. One emphasises as well as the word “likely”, the words “in the circumstances existing at the time”.

These pitfalls suggest that the reliability test could evolve into a predominantly objective standard, with judges admitting statements they deem actually reliable despite evidence of involuntariness. Even if not expressly stated, a judge might be influenced by the thought that the accused may have spoken the truth after all and thus such evidence should be admitted. Such a development would further entrench the already strict and objective character of Singapore’s law on admissibility. Accordingly, altering the voluntariness test’s foundation is unwarranted, not only because the reliability test’s utility is doubtful, but also because of the serious problems inherent in its application. First, the common law exclusionary rules in both jurisdictions differ from Section 122(5) CPC. In the UK, confessions are excluded if obtained through fear of prejudice, hope of advantage, or oppression. Because the rule lacked an objective requirement, it was often applied too generously in favour of the accused. The criticisms directed at the UK’s approach therefore do not apply to Singapore’s voluntariness test, which incorporates both subjective and objective elements. Second, the reliability test does not necessarily address the concern of excluding involuntary confessions that may nonetheless be true. The test focuses not on the actual reliability of the statement but on the likely effect of inducements on reliability. In practice, this produces little distinction from the voluntariness test, since both are objective inquiries requiring the court to assess the probable impact of interrogation. Judges cannot realistically determine abstract 68

R v Cox (1991) 93 Cr App R 287 (CA) 291.

207

Singapore Comparative Law Review

Unbroken Will

questions such as whether an accused spoke the truth despite coercion or inducement. Moreover, because both tests involve fact-specific evaluations, judicial assessments of interrogation effects will continue to vary widely. Third, adopting the reliability test risks negative consequences. Judges may be influenced by their own views of a statement’s actual reliability, thereby conflating admissibility with evidential weight. This undermines the protective principle underpinning the voluntariness test and could result in the admission of statements deemed actually reliable even though they were obtained through threats or inducements. Such a development would further entrench the already strict and objective character of Singapore’s law on admissibility, weakening the safeguards provided by the current framework. V.

SAFEGUARDS IN SINGAPORE’S CRIMINAL JUSTICE SYSTEM

Furthermore, there are already a few notable safeguards that are already employed in Singapore’s criminal justice system that result in the UK’s reliability-focused approach being unnecessary in Singapore’s context. A. Conviction based on a confession Caution is warranted before incorporating the reliability principle into Section 122(5) CPC, since in Singapore an accused may be convicted solely on the basis of a confession. A conviction may rest entirely on a confession, even if retracted, provided the court is satisfied that it was made voluntarily and is true, with no requirement for corroboration. In Lee Chez Kee v Public Prosecutor, the SGCA clarified that a confession is inherently a statement made directly against the maker's own interest, rendering it highly probative and reliable. 69 The court confirmed that a person may be convicted based on their pre-trial confession alone, without the legal necessity for any corroborative evidence. 70 In the UK, conviction on confession evidence alone is also possible. The Royal Commission on Criminal Justice endorsed this position, though only after considerable debate. 71 The majority of the Commission concluded that supporting evidence was unnecessary where the confession was credible and satisfied the requirements of PACE.72 However, they emphasised that juries should be strongly cautioned

Lee Chez Kee v Public Prosecutor [2008] 4 SLR(R) 20. ibid. 71 Royal Commission on Criminal Justice, Report (Cm 2263, 1993) 59. 72 ibid. 70

208

Singapore Comparative Law Review

Unbroken Will

about the dangers of convicting solely on confession evidence and directed to consider any available corroborating evidence. 73 The risk is even greater in Singapore because capital punishment remains in force for certain offences. For example, trafficking of certain types of prohibited drugs above a legislatively specified amount attracts the mandatory death penalty in Singapore. By contrast, the UK abolished the death penalty a substantial time ago. This highlights the need for Singapore’s exclusionary rules on confessions to provide robust protection for accused persons. If the system in Singapore is reformulated in terms of reliability, the ease of admitting confessions would increase.74 Therefore, it is submitted that the voluntariness test should not be radically replaced by a reliability test. The concept of voluntariness is not fundamentally flawed, and reform in this direction may fail to resolve the alleged shortcomings of the current test while creating new difficulties. Above all, any reform without careful consideration of Singapore’s unique context, as compared to other jurisdictions, risks undesirable and grave consequences for accused persons. B. Judicial Discretion to Exclude The position in Singapore was first articulated in Cheng Swee Tiang v Public Prosecutor,75 a case before a three‑judge High Court panel. The case was decided by a 2:1 majority. The offence concerned an unlawful lottery transaction in which police officers induced the appellant to sell tickets. The majority, comprising Wee CJ and Chua J, endorsed the common law approach, affirming that the courts retain discretion to exclude evidence on grounds of unfairness to the accused. Crucially, this discretion was not confined to the unfair use of evidence at trial but extended to broader considerations of protecting the accused from unlawful invasions of liberty while balancing the state’s interest in securing relevant evidence. 76 The judges did not attempt a definitive reconciliation of these competing interests, leaving the matter to be resolved on a case-by-case basis. In dissent, Ambrose J rejected the existence of such discretion, reasoning that it was not provided for in the Evidence Act, a view consistent

ibid. Ho Hock Lai, 'The Concept of Voluntariness in the Law of Confessions' (1991) 3 SAcLJ 66; Ho Hock Lai, 'On the Obtaining and Admissibility of Incriminating Statements' [2016] Sing JLS 249. 75 Cheng Swee Tiang v Public Prosecutor (1964) MLJ 291. 76 ibid. 74

209

Singapore Comparative Law Review

Unbroken Will

with Section 2(2) EA.77 Later authority, such as Wong Keng Leong Rayney v Law Society of Singapore, reaffirmed the existence of a discretion to exclude. 78 The English decision in R v Sang subsequently narrowed the common law position. 79 The House of Lords held that all relevant evidence, including that obtained through entrapment, was admissible unless its prejudicial effect outweighed its probative value. The discretion to exclude was confined to two categories: (1) where prejudicial effect outweighed probative value, and (2) where voluntariness of admissions or confessions was in issue, particularly in relation to evidence obtained from the accused after the offence. 80 In Muhammad bin Kadar v Public Prosecutor,81 the Court of Appeal clarified the scope of judicial discretion in excluding evidence. The case involved two brothers charged with murder, where the admissibility of police statements was contested. The Court affirmed that discretion exists, but it is confined to excluding evidence whose prejudicial effect outweighs its probative value in the context of trial proceedings. Applying this principle, the Court held that the High Court should have excluded two statements made by the accused under Section 121 of the CPC (now Section 22 of the CPC), notwithstanding their apparent voluntariness and admissibility under Section 122(5) of the CPC (now Section 258(1) of the CPC). The statements were deemed unreliable due to serious procedural breaches: deliberate non‑compliance with statutory requirements and police general orders, irregularities in recording, and the accused’s compromised physical state from drug dependency. The Prosecution failed to discharge its burden of explaining these manifest irregularities, and discrepancies between the original recording and subsequent diary entries further undermined reliability. In summary, it is posited that the judicial discretion to exclude statements where their prejudicial effect outweighs their probative value is a cornerstone safeguard in Singapore’s evidentiary framework. It empowers judges to act as the bulwark against potential unfairness by ensuring that evidence admitted at trial does not distort the fact‑finding process or compromise the accused’s right to a fair hearing. This discretion is significant because it allows courts to filter out statements that, while technically admissible under the voluntariness test, are rendered unreliable by procedural irregularities, deliberate misconduct, or contextual

ibid. Wong Keng Leong Rayney v Law Society of Singapore [2006] 4 SLR(R) 934 at [64]. 79 R v Sang [1980] AC 402. 80 ibid. 81 Muhammad bin Kadar v Public Prosecutor [2011] 3 SLR 1205. 78

210

Singapore Comparative Law Review

Unbroken Will

vulnerabilities of the accused. In doing so, judges prevent the prosecution from benefiting from evidence that could unduly sway the fact-finder without contributing meaningfully to truth‑seeking. By exercising this discretion, the Judiciary reinforces the protective function of the exclusionary rule, shielding accused persons from the consequences of improper investigative practices. This safeguard upholds the disciplinary function, deterring law enforcement from disregarding statutory safeguards. It also indirectly secures reliability, since evidence excluded under this principle is typically compromised in its trustworthiness. Taken together, this discretion operates as a strong safeguard as it preserves the integrity of criminal proceedings. Should a statement be deemed as voluntary as there is insufficient proof shown that the statement was involuntary, this judicial discretion serves a goalkeeping function that balances the interests of the state and the individual. Should any prejudicial statements slip past the defences of Singapore’s admissibility regime, the Judiciary will serve as the final line of defence, ensuring that convictions of accused persons rest on evidence that is probative as assessed by robustly trained judges (who are legal professionals themselves). Due to such a safeguard being in place, there is minimal to no impetus to reform the current admissibility regime. VI.

CONCLUSION

Singapore’s admissibility regime, anchored in the voluntariness test under Section 258 CPC, reflects a deliberate normative choice favouring certainty, efficiency, and doctrinal clarity. The triangular functions of the exclusionary rule – reliability, protective, and disciplinary – are already embedded within the voluntariness framework. Exclusions for coercion and inducement serve to safeguard reliability, judicial discretion deters misconduct, and the overarching burden of proof ensures fairness. While PACE has its own benefits, attempts to graft a reliability-centred model in Singapore that is akin to PACE would risk unnecessary complexity and undermine investigative efficiency without demonstrable gains. In light of this, Singapore’s current approach remains sufficient as it balances evidential integrity with procedural efficiency, preserves judicial flexibility, and avoids doctrinal fragmentation. There are also several common law constructs which are present in Singapore’s criminal justice system that vitiate against the need for the UK’s reliability-centric approach. Modification of the voluntariness test in Singapore is therefore unnecessary, as the existing 211

Singapore Comparative Law Review

Unbroken Will

framework already secures the essential values of the common law exclusionary rule while maintaining the pragmatic strengths of Singapore’s criminal justice system.

212

A Home Grown Law Firm Par Excellence About The Firm Lee & Lee LLP was established in 1955 by Mr Lee Kuan Yew, Mr Dennis Lee Kim Yew and Mdm Kwa Geok Choo. Lee & Lee LLP is one of Singapore’s leading law firms. Lee & Lee LLP remains committed to serving its clients' best interests, and continuing its tradition of excellence and integrity. The firm provides a comprehensive range of legal services to serve the differing needs of corporates, financial institutions and individuals.

Our Departments We have over 80 lawyers in the following 5 departments: • Banking | Loan & Other Debt-Related Transactions, Secured & Unsecured Financing and other banking matters • Corporate | Corporate & Commercial, Corporate Finance & Capital Markets, Mergers & Acquisitions, China Practice and Myanmar Practice

•

• •

Intellectual Property | IP Litigation, Prosecution & Enforcement, Brand Protection Strategies & Managing IP Portfolios, IP/IT Transactional Work, Technology, Media & Telecommunications, Privacy & Data Protection and Competition Litigation and Dispute Resolution | Civil Litigation, Criminal Litigation, Mediation and Arbitration

Real Estate | Sale & Purchase of Property, Collective Sale/Purchase of Property, Development Projects, Property Financing Transactions, Assignments and Debentures

Internship & Training We offer internships to 2nd year students for periods of 2 to 3 weeks in winter 2026. We also offer relevant legal training for 2027.

Contact If you are interested in interning or training with us, you may send your application to: (a) Mr Julian Tay (juliantay@leenlee.com.sg); and (b) Mr Matthew Saw (matthewsaw@leenlee.com.sg).

Join us today! Lee & Lee LLP 25 North Bridge Road Level 7 Singapore 179104 Tel: 6220-0666 | Fax: 6224-8505 | Email: leenlee@leenlee.com.sg

Visit our website

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

LAWFUL ACT DURESS: A COMPARATIVE STUDY OF ENGLAND AND SINGAPORE Chuan Wee Foo* I.

INTRODUCTION

Should lawful acts constitute duress? If so, where should the line be drawn? These questions engage one of contract law's most fundamental dilemmas: reconciling the demands of contractual certainty with the pursuit of substantive fairness. This article will seek to provide an answer to these two questions through a comparative lens, with reference to the jurisdictions of England and Singapore. Section II outlines the application of lawful act duress in England and Singapore. In Section III, it will be argued that both England and Singapore have, unlike other Commonwealth jurisdictions, rightly recognised lawful act duress as a vitiating factor and that the doctrine of lawful act duress can be normatively justified. In Section IV, this article will compare the differing approaches between the two jurisdictions before arguing that the Singaporean approach should be preferred. II.

LAWFUL ACT DURESS IN ENGLAND AND SINGAPORE A. England

In both England and Singapore, duress has long been recognised as a vitiating factor and if established, a victim is entitled to rescission. It is trite law that a threat to a victim’s person, property, and economic interests are valid grounds of duress.1 Much less certain was whether lawful acts could constitute duress in both jurisdictions. In England, this issue was squarely confronted by the UKSC in Times Travel (UK) Ltd v Pakistan International Airlines Corporation (“Times Travel”),2 which concerned a dispute over a contract entered between the claimant (Times Travel (UK) Ltd), a travel agent which sold flight tickets between UK and Pakistan, and the defendant (Pakistan International Airline Corporation (PIAC)), which operated the only airline flying directly between the UK and Pakistan. In response to the claimant seeking claims for unpaid commission, the defendant *

University of Cambridge, St Edmund's College, BA (Hons) In Law, Class of 2027. Hugh Beale, Chitty on Contracts (H G Beale gen ed, 36th edn, Sweet & Maxwell 2021) para 11-003. 2 [2021] UKSC 40. 1

213

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

threatened to significantly reduce the tickets allocated to the claimant unless the claimant agreed to enter into a new contract with the defendant – under which the claimant would agree to waive existing claims it might have for unpaid commission. Given that the claimant’s business would have been significantly affected if the defendant reduced its ticket allocation, the claimant relented and entered into a new agreement with the defendant. The claimant subsequently brought a claim to set aside the agreement on the ground of duress. Thus, the UKSC was invited to rule on whether a contract could be set aside for duress on the basis of lawful acts. Lord Hodge, delivering the majority opinion (Lord Reed, Lord Lloyd-Jones and Lord Kitchin agreeing), dismissed the appeal in favour of the defendants. Both the majority and Lord Burrows (who delivered a concurring opinion) explicitly recognised, for the first time, the existence in English law of the doctrine of lawful act duress.3 Although both the majority and Lord Burrows held that on the facts of the case lawful act duress was not established, the two camps had differing views as to when lawful act duress could be invoked. The majority held that lawful act duress would only be established in two instances (“Lord Hodge’s approach”): a) D uses his knowledge of criminal activity by C or a member of the claimant’s close family to obtain a personal benefit from the claimant by the express or implicit threat to report the crime or initiate a prosecution; or b) D, having exposed himself to a civil claim by the claimant, deliberately manoeuvres the claimant into a position of vulnerability by means which the law regards as illegitimate and thereby forces the claimant to waive his claim. 4 Lord Burrows opted for a wider approach and held that lawful act duress occurs where: (a) the threatening party has deliberately created, or increased, the threatened party’s vulnerability to the demand; and (b) the threatening party has made a demand in “bad faith” (a demand is made in bad faith where the threatening party does not genuinely believe that it has any defence and there is no defence to the claim being waived). 5

ibid [1]. ibid [4]. 5 ibid [136]. 4

214

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

B. Singapore Turning to Singapore, the issue of lawful act duress arose in Oon Swee Gek v Violet Oon Inc Pte Ltd (“Oon Swee Gek”). 6 In Oon Swee Gek, the claimants sought to set aside agreements made with the defendant on the basis that the contracts were vitiated on the grounds of economic duress and undue influence. For present purposes, the claim in economic duress is pertinent. The claimants alleged that the defendant had adopted commercially unfair practices in order to pressure them into entering different arrangements (“2019 Agreements”). These practices included: (i) threats to sue the claimants and expose him and his family members to claims on personal guarantees; (ii) exaggerating what the claimants had been “overpaid” to coerce the claimants into entering into the 2019 Agreements; (iii) exaggerating the defendant’s own contributions to convince the claimant to enter into the 2019 Agreements; and (iv) deliberately isolating the claimants in the course of negotiations. 7 Through these four tactics, the defendant sought to intimidate and confuse the claimants, thereby exerting pressure on them to enter into the 2019 Agreements, from which the defendant derived “significant benefits”. 8 In determining whether the defendant’s conduct amounted to economic duress, Jeyaretnam J reviewed the authorities on economic duress before summarising that economic duress could be established by the party seeking to rely on it if it could prove that: a)

The other party had exerted pressure directed at compulsion of their will

Such pressure was illegitimate, which entails an objective evaluation of the

pressure exerted and the overall circumstances. Mere commercial pressure is insufficient but it is not necessary that the pressure involve unlawful means. c)

But for the legitimate pressure, they would not have agreed at all or on those

terms.9 Therefore, Oon Swee Gek is authority for the proposition that lawful acts can constitute duress. In determining when lawful acts can constitute illegitimate pressure, Jeyaretnam J appeared to express support for the approach taken in Tam Tak Chuen v Khairul bin Abdul Rahman (“Tam

[2024] SGHC 13, [2024] 6 SLR 313. ibid [63]. 8 ibid [83], [86]. 9 ibid [60]. 7

215

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

Tak Chuen”).10 In Tam Tak Chuen, Prakash J (as she then was) cited the test in Nelson Enonchong, Duress, Undue Influence and Unconscionable Dealing (Sweet & Maxwell, 2006), holding that a threat of lawful action is illegitimate: 11 a)

Where the threat is an abuse of legal process

Where the demand is not made bona fide

Where the demand is unreasonable; and

Where the threat is considered unconscionable in the light of all circumstances

It is submitted that these factors (“Enonchong Factors”), though not overtly applied in Oon Swee Gek, were nonetheless influential in the finding that there was economic duress as part of Jeyaretnam J’s objective evaluation of the pressure exerted and the overall circumstances. This is reflected in the treatment of the defendant’s argument that the threats to sue the claimant and/or wind up the Company did not amount to illegitimate pressure as it was lawful, being neither a criminal nor civil wrong. In response to the latter argument, Jeyaretnam J held that such an argument was “too simplistic” as it overlooked the fact that the defendant was implicitly threatening legal action that he knew had no merit. 12 Thus, regardless of whether the defendant’s demands were lawful, they were, at minimum, unreasonable and not made in good faith.13 As such, the defendant’s conduct had moved beyond “mere commercial pressure” and could be regarded as “attempted exploitation”. 14 Therefore, on the facts, economic duress was established. C. Conclusion of Part II In sum, the conditions for establishing lawful act duress differ between England and Singapore. England has confined lawful act duress to two specific scenarios per Lord Hodge’s approach while Singapore takes a wider view per the Enonchong Factors, where bad faith and unreasonableness are relevant. III.

SHOULD LAWFUL ACT DURESS BE ALLOWED?

[2008] SGHC 242, [2009] 2 SLR(R) 240. ibid [50]. 12 Oon Swee Gek (n 5) [68] 13 Ibid [83] 14 Ibid. 11

216

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

A. Normative justification for lawful act duress Before comparing the two approaches, it is worth asking whether, in the first place, lawful acts should be able to constitute duress. Other Commonwealth jurisdictions such as Australia have rejected the concept of lawful act duress, confining duress to threats of or actual unlawful conduct.15 However, it is submitted that there are good reasons for the existence of Lawful Act Duress and that much of the criticism levelled against lawful act duress is overstated. First, it has been argued that lawful act duress undermines freedom of contract. The principle of freedom of contract includes the right to contract as one sees fit and conversely, the right to refuse to contract. However, lawful act duress opens up the possibility that a threat not to contract could constitute duress and thereby infringe upon party autonomy (eg: where a party in an existing contractual relationship with another threatens not to continue to contract unless a certain demand is met). In response, it is submitted that this argument should be rejected for two reasons. First, lawful act duress does not render refusing to contract unlawful, rather, it seeks to limit the circumstances where parties can use a refusal to contract to force another party into an accompanying demand. Second, even if such negotiation tactics are regarded as common in the course of commerce, it does not necessarily follow that particularly egregious forms of threats not to contract (ie: where such threats were made in bad faith) should be lawful.16 Even if lawful act duress leads to an inroad into freedom of contract, freedom of contract is not an absolute value and contract law has moved away from complete freedom of contract for practical or moral reasons (eg: the law on penalty clauses). 17 In light of the above, the argument that lawful act duress should not be allowed on the basis that it undermines freedom of contract should be rejected. An alternative argument that is often made against the imposition of lawful act duress would be that it is impossible to establish proper guardrails for its existence without it excessively undermining contractual certainty. Davies and Day posit that lawful act duress is an inherently “open-ended doctrine” that cannot exist without concepts such as “bad faith” (which they view 15

In ANZ Banking Group Ltd v Karam (2005) 64 NSWLR 149, the New South Wales Court of Appeal rejected the doctrine of lawful act duress, holding at [66] that the doctrine of duress should be confined to threatened or actual unlawful conduct. 16 An example of this is where “the accompanying demand seeks a sum of money the purpose of which is not the honest pursuit or furtherance of the duressor’s business interests” (CTN Cash and Carry Ltd v Gallaher Ltd [1994] 4 All E.R. 714, 718). The court alluded to the possibility of lawful act duress being made out in a case where a distributor threatens to withdraw credit in order to receive a sum which they did not believe was due. 17 Iona Branford and Jodi Gardner, 'Reconceiving Wrongdoing in Lawful Act Duress' (2023) 139 Law Quarterly Review 629-653, 636.

217

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

as too uncertain).18 Likewise, Morgan argues that “no clear and workable boundaries” exist in the realm of lawful act duress and that any distinction will inevitably be contested in future litigation.19 However, should the fact that the precise contours of lawful act duress be difficult to define prevent the courts from granting potential victims relief? One does not need to look beyond the other vitiating factors such as unconscionability and undue influence for evidence that the courts have exercised considerable discretion in order to ensure that contracts adhere to certain standards of procedural fairness. Any legal test is bound to be subject to contestation and it is thus argued that lawful act duress is not as anomalous as suggested by Davies and Day and Morgan. In dealing with lawful act duress, the English and Singaporean courts have acted in the traditions of the common law by setting out general principles with a view to incremental development. Additionally, allowing lawful act duress would be consistent with the broader doctrine of vitiating factors. When one considers lawful act duress in the context of other vitiating factors, more similarities than differences emerge. The common thread that runs through vitiating factors such as undue influence, duress, and unconscionability lies in the defect of the victim’s consent – in these cases, the victim’s consent has been procured through illegitimate means and is thus invalid. Thus, lawful act duress can be justified on the basis that lawful actions can, in some certain and limited circumstances, be regarded as illegitimate means of procuring consent. This argument is most persuasively conveyed when one considers the alternative – that lawful acts can never be illegitimate. The latter seems untenable when one considers the facts of Oon Swee Gek, where the defendant threatened legal action which he knew had no merit to exploit the claimant’s fear that he and his family might be exposed to personal guarantees. The defendant in Oon Swee Gek also deliberately made several other false accusations in order to induce fear in the claimant and manipulate him into entering the agreement. Such behaviour cannot fairly be regarded as everyday commercial bargaining tactics and to hold otherwise would be to normalise and accept that parties are free to make false accusations to create fear and manipulate the counterparty into entering an agreement. Although it could be argued that those in the claimant’s position could have protected themselves by seeking legal advice, this does not accord with the practical reality of negotiations where parties may 18

Paul Davies and William Day, '"Lawful Act" Duress (Again)' (2020) 136 Law Quarterly Review 7, 10. Jonathan Morgan, 'The Thin End of the Wedge: Morality, Contract Law and Lawful Act Duress' (2022) 81(1) Cambridge Law Journal 17, 20. 19

218

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

be limited by considerations such as costs and time. Hence, to suggest that lawful acts can never constitute duress reflects a limited understanding of the nature of duress and is inconsistent with the broader doctrine of vitiating factors. B. A role to play for Undue Influence and Unconscionability? It has also been suggested that cases where lawful act duress could potentially operate are better dealt with by existing and established vitiating factors. Davies and Day argue that cases of lawful act duress are best seen as instances where the threat is part of a chain of events originating from unlawful conduct and can thus be dealt with using existing principles of duress (taking the facts of Times Travel as an example, while PIAC’s threat to end contractual relations was something it was entitled to do, it was part of a “single chain of events” which started with PIAC’s unlawful breach of contract and the objective in making the threat was that it be excused from the prior breach of contract). 20 This approach, they argue, will be less uncertain than recognising the existence of lawful act duress as it avoids relying on notions such as “bad faith”. However, it is contended that Davies and Day’s proposed test of the concept of a “single chain of events” seems as broad or possibly even broader than a judgment of “bad faith”. For instance, would there be a temporal limit to the “chain of events”? Additionally, if PIAC’s demand that it be excused from its breach of contract was just one of several new terms in a new agreement, would it still be regarded as part of a “single chain of events”? Therefore, it is unlikely that Davies and Day’s approach will bring about the benefits of preserving contractual certainty to the extent that they have argued. Thus, given that both approaches would likely have similar effects on contractual certainty, this article takes the view that the better approach would be to accept that lawful acts can constitute duress rather than seeking to stretch existing principles of duress (as suggested by Davies and Day) as doing so would have the effect of eliding the distinction between lawful and unlawful act duress by subsuming the former under the latter. Similarly, it has also been argued that cases of lawful act duress can be dealt with on the basis of undue influence and unconscionability, extending these doctrines when necessary. This has been the approach taken by the Court of Appeal in New South Wales in Australia and New Zealand Banking Group Ltd v Karam,21 where the concept of lawful act duress was rejected in favour of maintaining the existing doctrines of undue influence and unconscionability. 20 21

Paul S Davies and William Day, ‘"Lawful Act" Duress’ (2018) 134 Law Quarterly Review 5, 7, 10. [2005] NSWCA 344, (2006) 64 NSWLR 149.

219

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

However, it does not appear that this solution is any better than recognising a separate doctrine of lawful act duress as this would undoubtedly widen the doctrines of undue influence and unconscionability as well. This is the point made by McBride, who persuasively argues that “all the supposed advantages of abolishing the law on lawful act duress would be lost if we simply substituted unconscionability for lawful act duress as the ground of relief, in cases that previously would have been dealt with (and could only be dealt with) under lawful act duress”.22 He then adds: “If we can identify a ground of relief (and when it will apply) without having to use the language of unconscionability we should – even if, ultimately, that ground of relief is based on considerations of what is conscionable and unconscionable. An analogy: we could say that A will owe B a duty of care when it is ‘fair, just and reasonable’ that he should, but if we can identify determinate situations where A will owe B a duty of care we should do so, and not rest content with a generic statement as to when A will owe B a duty of care. This is so even if all of the determinate situations where A will owe B a duty of care are ultimately based on the courts’ perceptions as to when it will be ‘fair, just and reasonable’ to find that A owed B a duty of care (how could they not be? – imagine a court thinking that it will find that A owed B a duty of care when it is ‘unfair, unjust and unreasonable’ to do so). The rule of law demands that we state the law with as much precision as possible – with the result that even if the law on lawful act duress is based on considerations of unconscionability (something conceded by Lord Hodge at [22]-[23] of Times Travel), we would be failing the rule of law if we jettisoned that area of law and allowed the cases covered by it to be subsumed within the more generic ground of relief that unconscionability represents.”23 Specific to the Singapore context, the courts have also explicitly recognised a ‘narrow’ doctrine of unconscionability and rejected calls for a broader definition of unconscionability in BOM v BOK:24

Nicholas McBride, 'Pakistan International Airline Corporation v Times Travel (UK) Ltd' (McBride's Guides, March 2022) https://mcbridesguides.com/wp-content/uploads/2022/03/Pakistan-International-AirlineCorporation-v-Times-Travel.pdf accessed 5 July 2026. 23 ibid. 24 [2018] SGCA 83, [2019] 1 SLR 349.

220

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

“In our view, it is possible to modify the original elements of the narrow doctrine of unconscionability without necessarily descending down the slippery slope into what is, in substance, the broad doctrine of unconscionability. In summary, and at risk of oversimplification, the narrow doctrine of unconscionability applies in Singapore. To invoke the doctrine, the plaintiff has to show that he was suffering from an infirmity that the other party exploited in procuring the transaction. Upon the satisfaction of this requirement, the burden is on the defendant to demonstrate that the transaction was fair, just and reasonable. In this regard, while the successful invocation of the doctrine does not require a transaction at an undervalue or the lack of independent advice to the plaintiff, these are factors that the court will invariably consider in assessing whether the transaction was improvident.”25 C. Conclusion of Part III Hence, this article takes the view that there is sound justification for the doctrine of lawful act duress and thus England and Singapore have both taken the right step in recognising lawful act duress. However, as seen in Part II, both jurisdictions have adopted different tests in relation to lawful act duress. In Part IV, this article will evaluate the two approaches before arguing that the Singaporean approach should be preferred. IV.

EVALUATION OF THE ENGLISH AND SINGAPOREAN POSITION A. Narrow vs Wide – a reflection of underlying differences

From the outset, it should be noted that England takes a narrow approach to lawful act duress while Singapore has taken a wider approach to lawful act duress. Lord Hodge’s approach confines lawful act duress to two specific instances, distilled from the existing case law.26 On the other hand, Singapore has adopted a more general test, choosing to rely on concepts such as “bad faith” and “unconscionability”. This article argues that the respective positions vis-à-vis lawful act duress in both jurisdictions stem from differing views as to the emphasis that should be accorded to contractual certainty.

ibid [140], [142]. Although it was initially unclear in Times Travel whether lawful act duress is only allowed in the two specific instances outlined by Lord Hodge, it has subsequently been affirmed in The Law Debenture Trust Corporation plc v Ukraine [2023] UKSC 11 that it is unlikely that a claim of lawful act duress will succeed outside the two established categories. 26

221

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

First, this is evident from the reasoning in Times Travel itself, as Lord Hodge criticised Lord Burrows’ proposed approach (which references the concept of “bad faith”) as being too uncertain and thereby undermining contractual certainty. Looking beyond Times Travel and lawful act duress, it is clear that English courts place an almost unyielding emphasis on contractual certainty as compared to Singapore. One such example can be seen from the treatment of non-oral modification (NOM) clauses. In Charles Lim Teng Siang v Hong Choon Hau (“Charles Lim”),27 the SGCA deliberated, in obiter, as to whether the position adopted by the UKSC in Rock Advertising Ltd v MWB Business Exchange Centres Ltd (“Rock Advertising”)28 should be followed in Singapore. In rejecting the majority’s approach in Rock Advertising, the SGCA reasoned that the UKSC’s approach was “overly concerned” with contractual certainty. 29 Instead, the SGCA opted for a rebuttable presumption that in the absence of an agreement in writing, no contractual variation would be permitted. 30 In so doing, the Singaporean courts rightly allowed for an inroad into certainty in favour of more practical considerations such as collective party autonomy – after all, contractual variation is particularly useful to most contractual parties given the dynamic nature of economic activity. The position that Singaporean courts have generally placed less emphasis on contractual certainty than English courts is challenged by the decision in Ochroid Trading Ltd v Chua Siok Lui31, where the SGCA departed from the English position in Patel v Mirza32 (“Patel”) in favour of a less discretionary and thus more rigid test for the doctrine of illegality. The primary justification given by the SGCA was the need for certainty: “To summarise the present part of this judgment, whilst the decision in Patel does furnish much legal food for thought, there are, with respect, difficulties with the approach adopted by the majority. While it may well be the case that, in practice there will be little difference between the approach of the majority in Patel and that of this court in Ting Siew May (at least where similar fact situations are concerned), the majority in Patel have, with respect, introduced further uncertainty into the analytical process by superimposing an additional inquiry based on a “range of factors” test across the board to all situations of common law illegality. We find such an approach to be undesirable

[2021] SGCA 43, [2021] 2 SLR 153. [2018] 4 All ER 21. 29 Charles Lim (n 22) [50]. 30 ibid [61]. 31 [2018] 1 SLR 363, [2018] SGCA 5 . 32 [2016] UKSC 42. 28

222

Singapore Comparative Law Review

creates

Lawful Act Duress: A Comparative Study of England and Singapore

an unprincipled

distinction between

the

principles

which

apply

to statutory illegality and those which govern common law illegality. It is also unnecessary to achieve remedial justice in the Singapore context given the flexibility of the principles laid down in Ting Siew May, which would (if relevant) also allow restitutionary recovery at the second stage of the inquiry. We hence do not accept the broader approach based on a “range of factors” set out by the majority in Patel and the present law on the question of whether the contract is prohibited which arises at the first stage of the inquiry remains unchanged.” Although the decision in Ochroid might cast doubt on the finding that Singaporean courts have placed less emphasis on contractual certainty than English courts, it should be noted that Patel is an exception to the ‘certainty-centric’ attitude in English contract law. Indeed, this was recognised by the majority in Patel, which justified the approach taken in Patel on the basis that contracts tainted by illegality should not enjoy the same guarantee of certainty that exists for lawful contracts: “Even if the claim to predictability of outcome for the reliance test could be made good, however, it is questionable whether particular weight should be given to this consideration in circumstances where a claimant and defendant have been parties to an agreement which is plainly illegal. Certainty or predictability of outcome may be a laudable aim for those who seek the law’s resolution of genuine, honest disputes. It is not a premium to which those engaged in disreputable conduct can claim automatic entitlement. For the reasons I have given, however, I do not believe that outcomes are easier to forecast on a rule-based approach” Considered in this light, Patel can be said to be an exception to the strict adherence to contractual certainty by English courts and it is thus fair to conclude that Singaporean courts are generally more willing to compromise on contractual certainty in favour of considerations of substantive fairness. Therefore, it is not surprising that Singaporean courts have adopted a broader approach to lawful act duress. B. Which approach should be preferred? It is submitted that the Singaporean approach should be preferred. It was evident in Times Travel that Lord Hodge opted for a narrow approach premised on the need to uphold the enforceability of contracts and contractual certainty. In response, this article takes the view that the English courts, in adopting a narrow approach, have placed excessive emphasis on contractual certainty. 223

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

First, a narrow approach premised on the need for contractual certainty sits uneasily with the normative justification underpinning the doctrine of duress. In essence, it prioritises contractual enforceability before considering whether agreement was validly obtained in the first place. As Phang argues, contractual enforceability must always be subsidiary to the broader question of economic duress: “This is, in part, to avoid a vicious circularity in reasoning where it might be argued that there is no illegitimate pressure by the mere fact of the existence of the compromise agreement itself; such reasoning clearly does not address the more important issue, which is whether the compromise agreement was itself the result of illegitimate pressure, which pressure could be the result of factors other than the fact of the compromise arrangement itself.” 33 Second, the strict adherence to contractual certainty by English courts overstates the importance of contractual certainty and understates the importance of substantive fairness. Contractual certainty has often (and rightly so) been seen as a core tenet of contract law. However, an excessive focus on contractual certainty prevents one from seeing the bigger picture. Phang astutely notes: “…the rules and principles which constitute the doctrine of the law are not ends in themselves but are, rather, the means through which the courts arrive at substantively fair outcomes in the cases before them in every area of the law”34 Therefore, although contractual certainty does have an inherent value, its importance, as a general rule of contract law, lies in it being a means to ensuring substantive fairness. A wider approach, such as that adopted by the Singaporean courts, can be justified based on the need to arrive at a substantively fair outcome. If we view contract law as a necessary framework to complement economic activity, it would not be a leap of logic to suggest that as much as parties expect that contracts will be sufficiently certain to conduct economic activity, parties also reasonably expect a degree of protection by the law from exploitation. This should not be taken to mean that the law should intervene to allow parties to escape from ‘bad bargains’ – lawful act duress is, after all, concerned with the conduct of parties prior to the formation of the contract and not the terms of the contract. As such, the wider approach taken by the Singaporean courts towards lawful act duress should be preferred as it provides courts with

33 34

Andrew Phang, 'Economic Duress: Recent Difficulties and Possible Alternatives' [1997] RLR 53, 60. Andrew Phang, 'Doctrine and Fairness in the Law of Contract' (2009) 29(4) Legal Studies 534, 535.

224

Singapore Comparative Law Review

Lawful Act Duress: A Comparative Study of England and Singapore

greater latitude to intervene for the purposes of substantive fairness. Further, the Singaporean courts have also affirmed that lawful act duress would only be established in “very rare” 35 circumstances and it follows that the bar for any successful claim of lawful act duress is accordingly high. Thus, even though the approach taken in Singapore is wider than the approach taken in English law, care has been taken to ensure that the inroads into contractual certainty are mitigated. V.

CONCLUSION

In conclusion, this article has canvassed the treatment of lawful act duress in England and Singapore. The recognition, in both jurisdictions, that lawful acts can constitute duress should be welcomed. However, England has adopted a narrow approach to lawful act duress while Singapore has adopted a wider approach to lawful act duress. Comparing the two perspectives, it is argued that the position in Singapore should be preferred as it places a greater emphasis on the nature of duress as well as the need for substantive fairness to be achieved.

Oon Swee Gek (n 6) at [53].

225

A different way to practise. Some firms handle cases. We handle crises. Setia is a specialist dispute firm, built for matters where the stakes are high, and the answers rarely straightforward. Led by Senior Counsel, our firm has been recognised in Chambers & Partners, Legal 500, Law.com, Lexology, Benchmark Litigation, and Global Investigations Review 100. While we’re known for the quality of our work, we’re equally known for the calibre of our people. We hold ourselves to the core values of loyalty and integrity—not just in what we do, but in how we go about it. Take it from our young lawyers: “the team creates a safe space for juniors...”, “the warm culture of the team extends beyond working hours”, “I feel like a valued member of a team rather than a mere cog in the machine.” What we offer: Complex, cross-border commercial mandates Tailored, thoughtful mentorship Meaningful and early exposure to advocacy Market-leading compensation and benefits Keen to find out more? Visit www.setialaw.com/careers/.

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

IN SEARCH OF MEANING: THE ECONOMIC TORTS Ho Kai En* and Parth Anand** Honourable Mention for Best Article Award 2026 I.

INTRODUCTION

The economic torts present an apparent paradox. A trader may lawfully harm a rival through competition, even deliberately, yet may face liability when that same harm is caused through breach of contract. The law’s difficulty is therefore not that economic loss may be intentionally caused. In competitive markets, it routinely is and is generally permitted by law. The real difficulty lies with identifying the point at which intentional economic harm becomes or should become legally wrongful. As courts move between restraint and purposive expansion, this movement reflects the difficulty of identifying a single clear rationale across torts that protect different interests in different contexts. The modern law is commonly grouped under the label “economic torts”, but that label conceals as much as it reveals. 1

The tort of inducing breach of contract protects contractual rights through accessorial

liability.2 The unlawful means tort imposes primary liability for targeted interference with economic relations through a third party. “Combination” and “unlawfulness” are technical elements of particular conspiracy torts, not a compressed description of one general wrong. Lawful means conspiracy is more anomalous still: otherwise lawful conduct may be actionable where the conspirators’ predominant purpose is to injure the claimant. 3 These are related liabilities, but they are not applications of one master rule. Academic accounts differ on how the category should be organised. Carty groups the economic torts around the protection of "financial expectations",4 while Blom and Burns cast the net wider to include breach of confidence, breach of fiduciary duty, and negligence. 5 No list *

University College London (UCL) (LLB Class of 2026) University College London (UCL) (LLB Class of 2026) 1 Clerk & Lindsell on Torts (24th edn, Sweet & Maxwell 2025), ch 23; Phang and Tan, Economic Torts in Singapore (Academy Publishing 2026) ch 2 2 OBG Ltd v Allan [2007] UKHL 21, [2008] 1 AC 1 3 OBG Ltd v Allan [2007] UKHL 21, [2008] 1 AC 1 [6]-[9], [32], [38]; Revenue and Customs Commissioners v Total Network SL [2008] UKHL 19, [2008] 1 AC 1174 [216], [225]; Crofter Hand Woven Harris Tweed Co Ltd v Veitch [1942] AC 435, 445-46, 462 4 Carty, An Analysis of the Economic Torts (Oxford University Press, 2nd Edn, 2010) at p 1 5 Joost Blom Peter Burns, Economic Torts in Canada (LexisNexis, 3rd Ed, 2024) at p 2. **

226

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

commands consensus. 6 Some accounts extend the label to deceit, passing off, malicious falsehood, breach of confidence or negligence; others confine it to the intentional interference torts. The most recent Singapore text reaches two conclusions that should frame the inquiry. First, the torts address the broad question of how far one person may harm another’s financial position in competitive activity. Secondly, each tort has its own “individual function, theory and rationale”, and full integration into a single framework may not presently be possible. 7 This essay argues that the economic torts do not yield a single clear rationale. The torts have accumulated too many historically specific and normatively distinct meanings to be reduced to one. Their apparent disorder reflects deeper tensions between freedom of competition, the protection of contractual rights, the security of commercial relationships, the control of collective pressure, and the common law's reluctance to recognise a general tort of unfair competition. The distinction matters especially in Singapore. Singapore courts are not merely receiving a neutral and complete English scheme. They must decide which historically contingent English compromises should be retained, narrowed or reformulated in a commercial setting where the reported cases have largely concerned companies, investments, shipping and complex transactions rather than the industrial conflicts that shaped much of the English law. The preferable method is therefore disciplined differentiation: identify the interest protected, the form of interference, the mental element and the adjacent legal regime for each tort, while coordinating the torts where their concepts genuinely overlap. The argument proceeds as follows. First, it distinguishes doctrinal coherence from functional coherence. Second, it examines the leading judicial account of unlawful interference through third parties in OBG v Allan. Third, it shows that OBG and Total Network did not resolve the field, instead it exposed persistent fractures in structure, intention, unlawful means, and party configuration. Fourth, it considers the pressure placed on the torts by competition law and labour or employment contexts. Fifth, it argues that Singapore should approach the economic

Lee Eng Beng, “A Perspective on the Economic Torts” [1996] SJLS 482; Clerk & Lindsell, paras 23-01 and 23-14 7 Phang and Tan, Economic Torts in Singapore paras 2.8-2.11, 2.40-2.43.

227

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

torts as a plural field, not as a doctrine waiting to be unified. The final part concludes that the field's coherence, if any, lies in disciplined differentiation rather than unity. II.

DOCTRINAL AND FUNCTIONAL COHERENCE

The question whether the economic torts are “coherent” can refer to two different aspects. A useful starting point is to distinguish doctrinal coherence from functional coherence. The former is presented as a structural harmony of legal elements across the economic torts - such that concepts like intention and unlawful means operate consistently. The latter, in alignment with Raz’s view that “what is coherent … makes sense … with all its bits hanging together”, 8 asks whether a tort considered in its practical operation serves a discernible normative rationale. Functional coherence does not presuppose that the entire field serves one purpose; it permits the possibility that different torts protect different interests while remaining intelligible within their own domains. In other words, doctrinal coherence concerns the fit between legal rules: whether concepts such as intention, unlawful means, justification and causation operate consistently across the torts. The literature often moves between these levels without identifying the move, which makes claims of either unity or incoherence appear stronger than they are.9 The scope of the category must also be fixed. In a broad sense, “economic torts” may include every tort or equitable wrong capable of protecting wealth or commercial expectations. That category is plainly too heterogeneous to bear one rationale. This essay therefore uses the narrower category of inducing breach, intimidation, causing loss by unlawful means and the two conspiracy torts. Even that narrower category is contested, but it presents the harder question because its members share intentionality, economic injury and a common history of policing interference with trade, contract and livelihood.10 On that question, the answer is twofold. The core torts can be understood as functionally coherent at a high level. Deakin and Randall provide the strongest account: the torts maintain the integrity of the competitive process by identifying protected economic interests, illegitimate interferences and available justifications.11 Their argument does not equate market regulation

J Raz, “The relevance of coherence”, in Ethics in the Public Domain (Oxford: OUP, 1994) 280 Simon Deakin and John Randall, ‘Rethinking the Economic Torts’ (2009) 72 MLR 519, 520, 531-33, 552-53; Clerk & Lindsell paras 23-01, 23-14-23-17; Phang and Tan, Economic Torts in Singapore paras 2.1, 2.40-2.42. 10 Clerk & Lindsell paras 23-01-23-02; EFT Holdings, Inc v Marinteknik Shipbuilders (S) Pte Ltd [2013] SGCA 64, [2014] 1 SLR 860 [70]-[73]. 11 Deakin and Randall, ‘Rethinking the Economic Torts’ 531-33, 552-53 9

228

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

with an open-ended judicial power to condemn “unfair” competition. Rather, it treats contractual protection, unlawful means and targeted interference as established legal devices for setting limits to rivalrous behaviour. 12 Phang and Tan similarly identify a central question - how far economic harm should be permitted—and present competition as the starting point unless there is a good legal reason to intervene. 13 That thin coherence is real and useful. It explains why intention is generally more demanding than foreseeability, why the courts resist a general tort of unfair competition, and why the torts must be interpreted with sensitivity to commercial freedom. It also explains why the separate liabilities should not develop in isolation. As Clerk & Lindsell observes, it would be as mistaken to pursue complete unity as to allow each tort to evolve without reference to the others.14 Thin functional coherence, however, is not a single justificatory rationale. A description such as “regulating competition” does not determine whether contractual rights deserve special third-party protection, why a wrong against an intermediary must be actionable, why criminal conduct suffices for conspiracy but not for the unlawful means tort, or why combination can make otherwise lawful conduct actionable. Those questions concern the nature of the wrong and the proper limits of liability. The broad function is compatible with several answers, each reflecting a different balance between security of contract, freedom of action, relational liberty, collective pressure and institutional restraint. The thesis is therefore not that functional coherence “does not matter” because doctrinal coherence is absent. Functional coherence supplies orientation and a vocabulary for comparison. The narrower claim is that it cannot do the adjudicative work of a thick rationale unless one moves from an account of the law as it stands to a programme of reform. Deakin and Randall themselves acknowledge that there is no uniformity after OBG v Allan and Total Network, and their proposed map seeks greater consistency rather than the merger of the existing heads of liability. 15 Phang and Tan make the same point more expressly: a fully consistent framework may require moving from what the law is to what it ought to be. 16 This distinction between explanation and reform is central to the argument that follows.

Deakin and Randall, ‘Rethinking the Economic Torts’ 531-33, 552-53. Phang and Tan, Economic Torts in Singapore paras 2.7-2.11 14 Clerk & Lindsell ch 23. 15 Deakin and Randall, ‘Rethinking the Economic Torts’ 551-53 16 Phang and Tan, Economic Torts in Singapore paras 2.40-2.43 13

229

Singapore Comparative Law Review

III.

In Search Of Meaning: The Economic Torts

LINDLEY/HOFFMANN RATIONALE: WAS THIS A CLEAR SINGLE RATIONALE?

This section examines the strongest judicial attempt to rationalise these torts, albeit one limited to the unlawful means torts. The closest judicial account to a unifying idea within the economic torts begins with Lord Lindley’s dictum in Quinn v Leathem.17 Ordinarily, a person who suffers because the defendant wrongs someone else cannot sue: the damage is too remote and the wrong belongs to the immediate victim. Lord Lindley reasoned that the position changes when the defendant wrongfully interferes with another’s liberty to deal with the claimant in order to injure the claimant. Targeted intention connects the third-party wrong to the claimant’s loss and defeats the objection of remoteness.18 Lord Hoffmann developed that insight in OBG v Allan. He separated inducing breach of contract from causing loss by unlawful means. For the latter tort, the defendant must intend to cause the claimant loss by acts that are unlawful as against a third party and that interfere with that third party’s freedom to deal with the claimant. The unlawful act must normally be actionable by the third party, subject to the qualification that the absence of loss to the third party does not prevent the act from qualifying. 19 The Supreme Court later confirmed in Secretary of State for Health v Servier Laboratories Ltd that this “dealing requirement” is an essential element of the tort. 20 This framework provides a conceptual structure through which we shall illustrate the doctrinal incoherencies that continue to subsist across the economic torts. The Lindley/Hoffmann rationale for the unlawful means torts rested on the dual foundations of the “right to trade” and “remoteness” theories. The former concerns a defendant who unlawfully interferes with a claimant’s liberty to deal with third parties; while the latter theory holds that a defendant’s unlawful act against a third party can give rise to a cause of action if the legal wrong is not too remote to reach the claimant where the defendant had intended to injure the claimant by means of it.21

Quinn v Leathem [1901] AC 495, 534–35 Quinn v Leathem [1901] AC 495, 534-35 (Lord Lindley); Philip Sales and Daniel Stilitz, ‘Intentional Infliction of Harm by Unlawful Means’ (1999) 115 LQR 411, 412-14. 19 OBG Ltd v Allan [2008] 1 AC 1 [45]-[56] (Lord Hoffmann) 20 Secretary of State for Health v Servier Laboratories Ltd [2021] UKSC 24, [2022] AC 959 [2], [63]-[74] 21 Neyers, J. (2008), Rights-based justifications for the tort of unlawful interference with economic relations. Legal Studies, 28, 221-225 18

230

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

The account is attractive within its proper domain. The claimant has no freestanding right to be insulated from competition. What the law protects is a more modest relational liberty: the opportunity to deal with others without the defendant intentionally disabling or coercing them through an independently wrongful act. The defendant’s intention to strike the claimant through the intermediary explains why the claimant, rather than only the intermediary, may recover. Rights-based accounts therefore defend the tort as protection of a claimant’s economic liberty against targeted wrongful interference, not as protection against economic loss as such. Firstly, it is submitted that the existence of the unlawful means tort rests on a questionable right. The “right to trade” postulates a duty upon defendants not to interfere, while according a negative liberty upon claimants. However, it is submitted that this conception was misplaced. This rationale appears to elevate a commercial privilege - the interest in trading free from interference - to the status of a protected legal right. In a competitive market economy, interference with others’ commercial interests is inevitable. Competition necessarily involves interference with others’ commercial interests. The common law has consistently refused to recognise a general tort of unfair competition or regulating competition through standards as open-ended as “unfairness”. To recast a trader’s interest in being free from competitive interference as a legal right is therefore to transform a commercial liberty into an entitlement the common law has never generally accepted. Even if that foundation is accepted, the remoteness theory does not clearly explain its proximity element. This leaves two questions unanswered: (1) why does the law require a specific intention to harm, rather than merely foreseeable harm; and 2) why may a claimant sue when the claimant’s own rights have not been directly violated. On the Lindley/Hoffmann account, the claimant may recover even where his own rights have not been directly violated, so long as the defendant intended to injure him by the unlawful interference with a third party. But that merely restates the doctrinal problem. It does not explain why targeted intention is required, nor why that requirement should be treated as the common organising principle across the torts. The theory identifies a control device; it does not supply a convincing normative foundation for why that control device should structure the field as a whole. These difficulties compound one another: a rationale that cannot explain why targeted intention is required within the threeparty structure it is designed to govern is even less equipped to account for torts that dispense with that structure altogether. Furthermore, this rationale cannot account for torts that do not depend on third-party interference. Even if the two party torts may have the same intention to

231

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

injure and use unlawful means, by not relying upon a third party interference, the Lindley/Hoffmann rationale is distinctly inapplicable to this subset. In the next section, we will examine the doctrinal incoherencies left in the wake of OBG and Total Network. Thus, the Lindley/Hoffmann framework, while influential in shaping the contours of the unlawful means tort, ultimately fails to provide a normatively consistent foundation for the economic torts as a group. It is against the backdrop of this failed rationale that the doctrinal incoherencies exposed in OBG v Allan and Total Network will be assessed. IV.

THE DOCTRINAL FRACTURES THAT DEFEAT MONISM

The absence of a singular rationale is clearest in four related fractures. These differences determine what must be proved, which interests receive protection and how liability is confined. A single rationale would have to explain why the law requires such different architectures. A. Accessory liability and primary liability The first fracture concerns the source of the defendant’s responsibility. In OBG, the House of Lords rejected the earlier tendency to subsume inducing breach of contract within a broader tort of unlawful interference with contractual or economic relations. The tort of inducing breach of contract is accessorial in character: it presupposes a valid contract, a breach by the contracting party, the defendant’s knowledge of the relevant obligation, and intentional procurement of the breach. The defendant’s liability is therefore parasitic on the contracting party’s primary wrong. Causing loss by unlawful means is different. The defendant is primarily liable for the defendant’s own conduct: using an unlawful act against a third party, interfering with the third party’s freedom to deal with the claimant, and intending thereby to cause the claimant loss. The claimant need not possess a contractual right against the third party, and the third party need not itself suffer loss. The wrong is the defendant’s targeted use of the intermediary, not participation in the intermediary’s breach. 22 This structural distinction is significant. It prevents the economic torts from expanding into a general liability for knowingly affecting another’s legal or economic interests. It also explains

OBG Ltd v Allan [2007] UKHL 21, [2008] 1 AC 1; Clerk & Lindsell on Torts (24th ed, 2025) paras 23-06, 23-08, 23-12

232

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

why mere interference with contractual performance, absent breach, no longer falls within Lumley v Gye after OBG v Allan. Such a claim must instead satisfy the separate and more restrictive requirements of the unlawful means tort, including the use of unlawful means and the requisite interference with the third party’s freedom to deal with the claimant. 23 The distinction further affects defences and corporate liability. A defendant who procures a breach of contract may, in appropriate circumstances, rely on justification or on the protective principle associated with Said v Butt, where a corporate officer acts bona fide within the scope of their authority..24 Singapore law recognises the same qualifications. In Bumi Armada Offshore Holdings Ltd v Tozzi Srl,25 the Court of Appeal held that factual inducement by a controlling shareholder is not by itself sufficient; an additional factor, such as lack of good faith or pursuit of an interest unrelated to the shareholder’s capacity, is required. Those limits are intelligible because inducing breach sits at the intersection of tort, contract and company law. These considerations do not translate straightforwardly to conspiracy or the unlawful means tort, where the focus falls on the defendant’s own unlawful conduct and the intentional causation of loss. Unlawful means conspiracy introduces a further complication. Total Network confirms that unlawful means conspiracy is not merely a route to imposing secondary liability where defendants jointly commit an independently actionable wrong. It is an independent tort. Once conspiracy is characterised in this way, and once its wrong is understood as intentionally causing injury by unlawful means, it begins to resemble the unlawful means tort. Its distinctive feature, however, remains the agreement or combination between defendants. The resulting overlap between primary liability, unlawful means, intention to injure, and combination is a significant source of the field’s continuing instability.26 A theory of intentional economic harm would tend to focus on the defendant’s purpose and the claimant’s loss. It does not explain why liability for inducing breach depends on the existence and breach of a particular contract, why corporate authority and shareholder capacity matter, or why an agreement between defendants can alter the result. The law’s distinct structures 23

OBG Ltd v Allan [2007] UKHL 21, [2008] 1 AC 1; Secretary of State for Health v Servier Laboratories Ltd [2021] UKSC 24; Clerk & Lindsell paras 23-08, 23-12 24 Said v Butt [1920] 3 KB 497 25 Bumi Armada Offshore Holdings Ltd v Tozzi Srl [2018] SGCA(I) 05; 26 Revenue and Customs Commissioners v Total Network SL [2008] UKHL 19, [2008] 1 AC 1174; Clerk & Lindsell paras 23-11, 23-14–23-17. Clerk & Lindsell notes both the independence of unlawful means conspiracy and the modern friction between the restrictive approach in OBG/Servier and the more flexible approach in Total Network

233

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

embody distinct judgments about the protected interest and the institutional boundary between tort and adjacent doctrines. B. The Meaning of “Unlawful Means” The second fracture concerns the inconsistent meaning of ‘unlawful means’, which further undermines doctrinal coherence. In OBG, unlawful means were defined narrowly as independently actionable wrongs against a third party, with criminal offences excluded unless they were also tortious. By contrast, Total Network held that unlawful means in conspiracy could include criminal offences, even where they were not actionable by a third party. This creates an anomaly, since unlawful means for conspiracy torts and for causing loss by unlawful means now diverge, despite appearing to share a conceptual core. Carty’s observation that Total Network “arguably undermined the prospect for clarity that OBG represented and threw the economic torts back into the mess in which they were before” 27 captures the importance of the divergence: if the concept of unlawful means cannot bear a consistent meaning even across torts that nominally share it as a common element, then the economic torts lack the shared conceptual core that a single rationale would require. Recent Singapore commentary confirms that this is not merely an English difficulty. In Singapore, the scope of “unlawful means” remains unsettled in both the unlawful means tort and unlawful means conspiracy. Beckkett endorsed the Total Network position that unlawful means conspiracy does not require the unlawful means to be independently actionable and may include criminal acts or intentional tortious acts. 28 EFT Holdings then accepted that important questions remained open, including whether lawful means conspiracy should continue to exist, the scope of unlawful means in unlawful means conspiracy, and the limits of liability where criminal conduct is relied upon. 29 The result is not a clean post-OBG settlement, but an unstable accommodation: the unlawful means tort remains tied to a narrower third-party dealing model, while unlawful means conspiracy has developed a more flexible and potentially wider conception of unlawfulness. Phang accordingly describes the position as unsettled and regards instrumentality as a possible alternative to actionability as the relevant control mechanism. 30

H Carty, 'The Economic Torts in the Twenty-First Century' (2008) 124 LQR 641, 642 Beckkett Pte Ltd v Deutsche Bank AG [2009] SGCA 18, [2009] 3 SLR(R) 452 [120]. 29 EFT Holdings [2014] 1 SLR 860 [85], [90]-[92]; Phang and Tan, Economic Torts in Singapore paras 5.125.14. 30 Phang and Tan, Economic Torts in Singapore paras 4.37-4.40, 5.52-5.54, 13.24 28

234

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

The pending UK Supreme Court appeal in Massa v Formula One Management Ltd makes this instability especially vivid. 31 At first instance, Jay J permitted the unlawful means conspiracy claim to survive summary judgment even though the alleged unlawful means included contractual duties allegedly owed by the FIA to its members rather than to Mr Massa himself. That reasoning treats unlawful means conspiracy as a primary tort whose unlawful means need not be independently actionable at the claimant’s suit. The Supreme Court has now granted permission to appeal on whether unlawful means conspiracy can be founded on a civil wrong not independently actionable by the claimant, a breach of contract to which the claimant is not party, a breach of foreign law, and conduct which the defendant did not know to be unlawful. The appeal therefore goes directly to the fractures identified in this essay: actionability, privity, foreign unlawfulness, and knowledge of unlawfulness. Until the appeal is decided, Massa should be treated not as a settled answer, but as evidence that the limits of unlawful means conspiracy remain doctrinally alive. C. Intention The intention requirement across the economic torts is conceptually fragmented. OBG provided some clarity by distinguishing intention to injure from mere foreseeability of harm, framing the analysis in terms of “ends” and “means”. 32 Yet this standard falters in complex scenarios, for example where a defendant’s gain-based conduct causes incidental loss to the claimant. 33 Edmundson expresses a concern in this tendency to allow “the elements of one tort to permeate analysis of another without rigorously addressing the differences” suggesting it causes the danger of a conceptual osmosis.34 This risk of elements of one tort permeating the analysis of another - without confronting the differences between them - is illustrated by Lord Hoffmann's folding of intimidation into unlawful interference, to the unexamined transplanting of the “unlawful means” concept across OBG and Total Network, ultimately depicting doctrinal incoherence. Further, while the Court of Appeal in EFT Holdings recognised that intention is a necessary element across the general economic torts, its content cannot be given a uniform meaning; 35 rather, it must be defined by reference to each tort's specific rationale. Thus, the pursuit of a 31

Massa v Formula One Management Ltd and others [2025] EWHC 3064 (KB); Massa and others v Formula One Management Ltd UKSC/2026/0035 32 OBG Ltd v Allan [2008] 1 AC 1 [42], [62], [135], [164]. 33 See W.H. Newson Holding Ltd v IMI Plc [2013] EWCA Civ 1377 34 Edmundson, P, ‘Conspiracy by unlawful means: Keeping the tort untangled’ 16 Torts LJ (2008) 189, 190 35 EFT Holdings at [85]

235

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

single rationale cannot be rescued by pointing to intention as a shared requirement when the content of that requirement is itself irreducibly plural. Singapore law also requires a more careful account of intention. EFT Holdings rejects the proposition that knowledge of likely harm is by itself sufficient. The claimant must show that injury was intended, either as an end in itself or as a necessary means to the defendant’s end.36 Singapore authority again illustrates that choice. In Crest Capital Asia Pte Ltd v OUE Lippo Healthcare Ltd, the Court of Appeal considered whether conspirators must know that the means deployed were unlawful. Although it was strictly unnecessary to determine the issue because actual knowledge had been established, the Court affirmed that knowledge of unlawfulness is not an element of unlawful means conspiracy, reasoning that intention to injure already supplies the principal control against excessive liability. 37 In Paragon Shipping Pte Ltd v Freight Connect (S) Pte Ltd, the High Court applied the three-party unlawful means formulation and rejected the claim because unlawfulness, targeted intention and causation were not established.38 This approach may be defensible, but it reinforces the point that the mental element is not uniform across the field (nor across jurisdictions noting Racing Partnership).39 This distinction is crucial because it prevents unlawful means conspiracy from collapsing into liability for commercial conduct carried out with awareness of adverse consequences. Yet it also shows that intention does different work across the economic torts: in inducing breach it fixes upon procurement of breach; in unlawful means it connects the defendant’s unlawful conduct to the claimant’s loss; in conspiracy it operates as the principal control device for collective intentional harm. D. Two-party and three-party structures The fourth fracture concerns party structure. The OBG unlawful means tort is fundamentally three-party: the defendant commits an unlawful act against a third party in order to harm the claimant. But Total Network involved conspiracy and direct harm. Carty observes that Total Network appeared to accept a broader space for two-party economic torts, or at least for direct intentional harm by unlawful means, while leaving the boundaries uncertain.

EFT Holdings at [99]–[101] Crest Capital Asia Pte Ltd v OUE Lippo Healthcare Ltd [2021] SGCA 25, [2021] 1 SLR 1337 [123]-[129]; Phang and Tan, Economic Torts in Singapore paras 5.67, 13.32 38 Paragon Shipping Pte Ltd v Freight Connect (S) Pte Ltd [2014] SGHC 165 [82]-[84]. 39 The Racing Partnership Ltd v Sports Information Services Ltd [2020] EWCA Civ 1300, [2021] Ch 233 at [141]–[144], [171], [265] 37

236

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

This distinction is not a technicality. Recognising a broad two-party tort of causing loss by unlawful means would risk converting crimes, regulatory breaches and statutory wrongs into private causes of action beyond what Parliament intended. Yet refusing any two-party claim may leave deliberate economic injury by unlawful means without a remedy. The common law has responded through piecemeal distinctions rather than through a unified theory. That piecemeal character is one reason the economic torts resist reduction to a single rationale. The same difficulty appears in the relationship between intimidation and the unlawful means tort. Three-party intimidation can be understood as a species of unlawful means liability because the defendant pressures a third party in order to harm the claimant. Two-party intimidation, however, concerns pressure applied directly to the claimant. It survives as a separate tort or at least as a separate structural problem. The existence of such borderline forms confirms that party configuration is not incidental. It is part of the architecture of liability. V.

LACK OF A SINGLE CLEAR RATIONALE: CAN A SINGLE CLEAR RATIONALE EVER EXIST?

It is submitted that even if the economic torts are doctrinally fragmented, they are redeemed by functional coherence. However, it is submitted that functional coherence at the level of particular torts does not yield, and cannot substitute for, a single clear rationale across the whole field. The economic torts - positioned at the intersection of commercial freedoms and labour rights - occupy an inherently contested area. The lack of a single rationale reflects these deeper normative disagreements. For example, as Lord Scott observed in Total Network, the unlawful means conspiracy tort was not the product of principled development, but of “stage by stage” judicial construction. 40 This piecemeal evolution is indicative not of a coherent rationale underpinning the economic torts, but of a lacuna in one: a doctrinal gap that continues to resist unified theoretical justification. On this point, Lord Nicholls in OBG acknowledged that even the meaning of “unlawful means” has no consensus, attributing the divergence to a deeper disagreement over the tort’s rationale. 41 Therefore, it may seem that each tort’s framework assumes a gap-filling role, to overcome the absence of a single clear rationale. A more ambitious alternative was proposed by Lee Eng Beng in 1996. Lee argued that the common law should recognise a general principle of liability for intentionally inflicted economic or non-physical harm without lawful cause or justification, accompanied by a broad 40 41

Total Network (n 3) [56] OBG (n 2) [152],[155]

237

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

and flexible defence of justification. 42 The defence would take account of the rights invoked by the defendant, violation of the claimant’s rights, unlawful means, malice and commercial unfairness.43 It would treat the existing torts as applications or qualifications of a prima facie wrong of intentional harm. Lee’s proposal remains normatively intelligible, but it should be recognised as reform rather than an explanation of current law. It requires turning away from Allen v Flood, which established that an otherwise lawful act does not become tortious merely because it is done with a bad motive. It would also reverse the direction of OBG and Servier, which favour sharply defined elements and manageable limits over open-ended judicial assessment of commercial fairness. A broad justification defence could yield morally attractive outcomes, but it would transfer to courts a substantial power to balance commercial interests case by case. That may reduce predictability, permit tort to bypass statutory competition or labour choices, and obscure the special reasons for protecting contracts or condemning combination. The recent Singapore book helps sharpen the answer to Lee. Its authors do not deny that a more consistent framework could be designed. They distinguish, however, between the current “kinks” in the law and the normative work required to remove them. Lee’s model is therefore one possible ought, not the hidden is of the economic torts.44 Singapore courts should not adopt such a general prima facie tort through the incremental resolution of cases without confronting its institutional and cross-doctrinal consequences. Lawful means conspiracy provides a final test. Nathan Tamblyn argues that the tort is best understood not as an economic tort but as an instance of abuse of rights in which malice converts the exercise of lawful rights into an abuse. 45 Whether or not that account is accepted, it exposes the difficulty of monism. If lawful means conspiracy is an abuse-of-rights doctrine, its rationale is distinct from accessorial protection of contract and third-party unlawful means. If it is instead a device for policing excessive competition or collective pressure, it overlaps uneasily with modern statutory regulation. Either characterisation supports coordination without conceptual merger.

Lee Eng Beng, ‘A Perspective on the Economic Torts’ [1996] Sing JLS 482 Allen v Flood [1898] AC 1, 154; OBG Ltd v Allan [2008] 1 AC 1 [35]-[38], [56]; Servier [2022] AC 959 [59]-[74]; 44 Lee Eng Beng, ‘A Perspective on the Economic Torts’ 522-23; Phang and Tan, Economic Torts in Singapore paras 2.42-2.43 45 Tamblyn, ‘Lawful Act Conspiracy’ 159-61 43

238

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

The appropriate conclusion is thus qualified. The economic torts have enough functional coherence to be studied together and to inform one another. They do not have a single existing rationale capable of generating their elements. Coherence should mean that differences are made explicit, justified and coordinated - not that they are forced into one formula. A central conflict exposing the absence of a single clear rationale is whether the economic torts serve corrective justice or perform a broader regulatory function. On a corrective justice view, the torts respond to wrongful interference with protected private interests, especially contractual rights and commercial relationships. On a broader regulatory view, they help determine the limits of permissible economic pressure in markets, labour disputes, and competitive activity. Lord Hoffmann’s framework reduces economic torts to the protection of two interests: the performance of valid contracts and the freedom of third parties to deal without interference. 46 This gives the law a clearer structure, but it also creates tension. If the economic torts merely police wrongful interference, they should not disturb ordinary risk allocation in a laissez-faire market. Yet in labour and commercial disputes, the same torts may operate to protect contractual expectations against collective pressure or competitive disruption. This tension is visible in the shape of the individual torts. The Lumley v Gye tort treats contractual rights as interests deserving heightened protection, while the unlawful means tort focuses on independently wrongful conduct that disrupts the claimant’s market relationships. Conspiracy introduces a further concern with combination, purpose, and the legitimacy of collective action. These different mechanisms reflect different judicial commitments: protecting bargains, preserving commercial liberty, and preventing excessive interference with economic relations. The boundaries of the economic torts are therefore not fixed by a single abstract principle, but are shaped by the socio-political context in which they operate. The result is a multiplicity of rationales pulling in different directions.

Bagshaw, Roderick, ‘Lord Hoffmann and the Economic Torts’ (2015). PS Davies and J Pila (eds), The Jurisprudence of Lord Hoffmann (Oxford: Hart Publishing, 2015), Oxford Legal Studies Research Paper No. 9/2015, 2-3

239

Singapore Comparative Law Review

VI.

In Search Of Meaning: The Economic Torts

LACK OF A SINGLE CLEAR RATIONALE: DISSONANCE OF FRAMEWORKS

It is submitted that the foundational absence of a single shared rationale coheres with the absence of a unified theory underpinning the economic torts - as affirmed in OBG -47 suggesting that coherence in purpose can never compensate for doctrinal disunity. Tort law mediates complex interactions between protected interests and permissible conduct. It is therefore unlikely that the economic torts can be reduced to conduct-based rules alone. 48 On this view, the absence of a unified doctrinal structure is not a shortcoming to be resolved, but a reflection of the pluralistic nature of tort law itself. 49 OBG’s rejection of the unified theory has shown judicial sensitivity to the conceptual justifications underpinning each tort rather than a willingness to transpose notions from one tort into another on the basis of supposed family resemblance alone. That move matters because once one accepts that Lumley v Gye is accessory liability while the unlawful means tort is primary liability, the prospect of one master principle becomes distinctly remote. The field is better understood as a cluster of torts that require coordination. Singapore’s post-OBG case law points in the same direction. In Bumi Armada, the Court of Appeal held that where a parent company induces its subsidiary to breach a contract, something more than factual inducement is needed before liability in tort arises; at least on the facts, bad faith or pursuit of an interest beyond the parent’s role as shareholder would be required. In EFT Holdings, the Court of Appeal insisted upon a strict, targeted conception of intention for unlawful means conspiracy. In Paragon Shipping, the High Court restated wrongful interference with trade in narrow OBG-style terms. These are not moves toward a general principle of intentional economic harm. They are illustrations of tort-specific limits developing separately, and for good reason. Therefore even if a residual role for a single rationale existed, the courts have actively moved away from monism and adopted a pragmatic, differentiated approach. The modern law is more likely to develop as a functionally intelligible but doctrinally non-unified body. That is not necessarily a failure. It may instead reflect the pluralistic structure of tort law itself, where

See OBG (n 2) [38],[264] OBG (n 2) [32] 49 Peter Cane “Mens Rea in Tort Law” (2000) 20 Oxford JLS 533, 552 48

240

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

different protected interests and different kinds of conduct properly call for different forms of liability. VII.

COMPETITION, LABOUR, AND THE LIMITS OF GAP-FILLING

The limits of a single rationale become sharper when the torts interact with distinct areas of law. Competition law presents the most obvious overlap. Claimants have sometimes pleaded unlawful means conspiracy or causing loss by unlawful means as alternative routes where a statutory competition claim is unavailable, difficult or confined by questions of standing and protected purpose.50 Carty likewise observes that modern claimants seek to expand the torts beyond their traditional role in trade conflict so that they fill gaps in liability for commercial wrongdoing.51 That overlap does not make the economic torts a general private enforcement mechanism for competition norms. The common law has not recognised an all-embracing tort of unfair competition. In Servier, the Supreme Court adopted a restrictive reading of the unlawful means tort partly because the regulation of economically sensitive activity should largely be left to Parliament and because the dealing requirement supplies manageable limits. 52 The doctrinal point follows directly: a broad market-regulation function cannot justify importing the wider concept of unlawfulness used in conspiracy into the three-party unlawful means tort, or treating every competition-law breach as tortious at the suit of every person who suffers loss. Each cause of action retains its own standing, mental-element and relational requirements. Lawful means conspiracy occupies a particularly uneasy position. Historically, it resembles a common-law response to collective economic pressure; modern competition legislation now performs much of the work of controlling collusion and market power. Yet the tort has not disappeared, because it can reach conduct or relationships outside statutory regimes. Its residual survival does not prove that the common law and competition law share one standard. It reflects a narrow judgment that concerted lawful action may exceptionally become wrongful when its predominant purpose is injury. 53

Grant Stirling, ‘It’s a Conspiracy! Or Is It? The Difficulty with the Economic Torts as “Alternative” Causes of Action for Competition Law Damages Actions in UK Courts’ (2017) 8 Journal of European Competition Law & Practice 233, 233-35 51 Hazel Carty, ‘The Modern Functions of the Economic Torts: Reviewing the English, Canadian, Australian, and New Zealand Positions’ (2015) 74 CLJ 261, 264-67 52 Servier [2022] AC 959 [59]-[74] 53 Tamblyn, ‘Lawful Act Conspiracy’ 159-61; Carty, An Analysis of the Economic Torts 145-47

241

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

Labour law pulls the same torts in a different direction. Their modern form was shaped by industrial conflict. In Allen v Flood, lawful pressure by union officials did not become actionable merely because it was malicious. In Quinn v Leathem, combination and an injurious purpose supported liability. In Crofter, the House of Lords treated the advancement of the defendants’ collective interests as inconsistent with a predominant purpose to injure. In Rookes v Barnard, threatened breaches of employment contracts supplied the unlawful pressure required for intimidation and conspiracy. 54 The cases are not simply applications of market rivalry. They concern the legitimacy of collective bargaining power, association and industrial action. The statutory response confirms the contextual character of the law. In the United Kingdom, Part V of the Trade Union and Labour Relations (Consolidation) Act 1992, particularly s 219, grants immunity from specified tort liabilities for acts done in contemplation or furtherance of a trade dispute, alongside detailed requirements concerning ballots and notice. 55 The practical legality of industrial action therefore cannot be read from the common-law elements alone. As Clerk & Lindsell stress, the economic torts’ role in labour relations must be understood against the statutory developments since 1875. 56 Competition and labour law therefore support thin functional coherence while defeating thick unity. Both fields ask when economic pressure becomes impermissible, but they answer through different institutional arrangements and policy balances. Market rivalry favours freedom to compete subject to statutory and established private wrongs. Labour law mediates contract, collective action and statutory immunity. A single general tort would flatten those differences. A differentiated family of torts can accommodate them, provided the reasons for each boundary are stated candidly. VIII. SINGAPORE: COHERENCE THROUGH DISCIPLINED DIFFERENTIATION The comparative inquiry is justified because Singapore has received English economic tort doctrine without fully reproducing the historical conditions that shaped it. English law developed against the background of labour conflict, trade union activity, and judicial 54

Allen v Flood [1898] AC 1; Quinn v Leathem [1901] AC 495; Crofter [1942] AC 435; Rookes v Barnard [1964] AC 1129, 1204-07 55 Trade Union and Labour Relations (Consolidation) Act 1992 56 Clerk & Lindsell paras 23-03, 23-17-23-18, 23-135 ff; David Howarth, ‘Against Lumley v Gye’ (2005) 68 MLR 195, 196-99

242

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

reluctance to create a general common-law doctrine of unfair competition. Those pressures help explain why English doctrine took the form of contested compromises rather than a single coherent theory. In Singapore, however, much of that background is absent or less central. The inherited doctrines therefore cannot simply rely on English judicial anxieties for their justification. What may appear in England as a historically explicable compromise becomes, in Singapore, a set of doctrinal fragments whose rationale must be supplied afresh. In this sense, the incoherence is not merely inherited; it is made more visible because Singapore courts must decide whether these English compromises should be preserved, narrowed, or reformulated for a different commercial and legal context. For Singapore, the comparative question is not whether English law contains a hidden unity waiting to be received. It does not. The question is which parts of the English compromise should be retained, narrowed, or reformulated. Singapore authority already points toward differentiation rather than unity. In EFT Holdings, the Court of Appeal articulated the elements of unlawful means conspiracy. In Crest Capital, it clarified that knowledge of unlawfulness is not presently required, because intention to injure serves as the central control mechanism. 57 In Bumi Armada, the Court of Appeal held that a shareholder or parent company acting in good faith in its capacity as shareholder should not ordinarily be liable for inducing the company’s breach of contract. 58 In Kapital Fund, the High Court summarised the Singapore position that directors are generally protected where they act within their authority and in accordance with duties owed to the company, but may lose that protection where they breach personal duties owed to the company.59 Recent Singapore commentary takes the same view. It emphasises that many questions remain unsettled in Singapore, including the scope of the Lumley v Gye tort, the unlawful means tort, the conspiracy torts, and the applicability of two-party and three-party structures. That uncertainty is not a weakness if it is openly acknowledged. It gives Singapore courts room to develop a law of economic torts that is coherent at the level of specific torts, without pretending that the field has a single foundation.

EFT Holdings [2014] 1 SLR 860 [70]-[72], [85]. Bumi Armada [2019] 1 SLR 10 [45]-[49]; Phang and Tan, Economic Torts in Singapore paras 3.17-3.18. 59 Kapital Fund SPC v Lee Tze Wee Andrew and another [2024] SGHC 289 58

243

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

Bumi Armada illustrates the point in inducing breach of contract. Where a parent company induces its subsidiary to breach a contract, the Court of Appeal held that something more than factual inducement is required before liability in tort arises. At least on the facts, bad faith or pursuit of an interest unrelated to the parent company's capacity as shareholder would be required. This is not an application of a general principle against intentional economic harm. It is a tort-specific limit shaped by company law, shareholder autonomy, and the policy behind the Said v Butt protection for corporate actors. 60 EFT Holdings and Crest Capital illustrate the same approach in conspiracy. EFT Holdings articulates a targeted conception of intention, rejecting liability based merely on awareness of likely harm. Crest Capital, while not finally deciding every point, suggests that intention to injure may control liability even where knowledge of unlawfulness is not required. 61 These are specific limiting devices, not moves toward a general prima facie liability for intentionally caused economic loss. This is why Lee's 1996 proposal for a general principle of liability for intentionally inflicted non-physical harm, subject to a broad defence of justification, should be resisted today. The proposal was understandable when the field appeared under-theorised. But later authority has moved in the opposite direction. OBG rejected the unified theory. Servier reaffirmed restrictive and manageable limits. Singapore's post-OBG case law likewise proceeds through specific control mechanisms rather than through a broad prima facie liability rule. Lawful means conspiracy confirms the point. If any tort resists assimilation into the ordinary economic torts, it is this one. Its liability turns not on unlawful means but on combination and predominant purpose. Tamblyn's view that lawful act conspiracy is better understood as an instance of abuse of rights rather than as an ordinary economic tort strengthens the argument that the field is plural rather than unified. 62 The unlawful means tort presents a further choice. Phang and Tan have argued for a wider category of unlawful means, including civil and criminal wrongs, controlled by a nuanced test of instrumentality. Their proposal deserves serious consideration because it confronts the arbitrariness of actionability and asks whether the third-party wrong was genuinely used as the mechanism of injury. It also illustrates the proper level of reform. Instrumentality is a proposed

Said v Butt [1920] 3 KB 497 EFT Holdings [2014] 1 SLR 860 [98]-[101], [111]; Crest Capital [2021] 1 SLR 1337 [123]-[129]. 62 Tamblyn, ‘Lawful Act Conspiracy’ 159-61 61

244

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

control for a particular tort and relational structure; it does not require “unlawful means” to have the same content in conspiracy. Four questions should guide Singapore courts. First, what interest is protected? Inducing breach protects the claimant’s contractual entitlement; the unlawful means tort protects an economic relationship against targeted third-party wrongdoing; conspiracy addresses concerted intentional injury. The interest should not be described at the empty level of “economic loss”. Secondly, what form of interference makes the conduct wrongful? Procurement of breach, instrumental wrongdoing against an intermediary, coercion and combination are distinct. The court should resist converting one into another merely because the facts are commercially unattractive. Thirdly, what mental element confines liability? The relevant object may be breach, loss or predominant injurious purpose. The ends-means-consequences framework is useful, but its application must remain tied to the wrong in question. Fourthly, which adjacent regime supplies or limits the policy? Contract law, company law, competition legislation, labour law and statutory causes of action make their own choices about rights, remedies and standing. Economic torts should not be used to bypass those choices without clear justification. This method preserves both certainty and development. It does not freeze Singapore law at OBG. The courts may prefer actionability, instrumentality, targeted intention, justification, causation or proof of loss as control mechanisms, depending on the tort. But they should identify the choice as tort-specific, compare it with neighbouring liabilities and explain its institutional consequences. That is the disciplined differentiation advocated here. Such an approach also answers the concern that pluralism licenses disorder. Coordination remains essential. Similar concepts should presumptively be interpreted consistently where they perform the same role, and departures should be reasoned. Yet formal consistency is not an end in itself. The same word may properly differ when it responds to a different wrong, just as a common function may be implemented through different doctrinal structures. The objective is a transparent and principled map, not a fictitious unity. IX.

CONCLUSION

The economic torts are coherent in one limited sense and incoherent in two stronger senses. They possess thin functional coherence because they address deliberate interference with economic interests and help define the permissible limits of competitive pressure. They lack doctrinal unity because inducing breach, causing loss by unlawful means, unlawful means conspiracy and lawful means conspiracy rest on different structures, meanings of unlawfulness, 245

Singapore Comparative Law Review

In Search Of Meaning: The Economic Torts

objects of intention and party configurations. They also lack a single rationale capable of explaining those differences and resolving hard cases without additional policy choices. That conclusion does not condemn the field as meaningless or merely “ramshackle”. It explains why the search for one master rule repeatedly fails. The protection of contractual rights, the extension of liability for targeted third-party wrongs, the control of collective unlawful action and the exceptional condemnation of predominantly injurious lawful action are distinct normative projects. Competition and labour law sharpen those differences, while OBG, Total Network, Servier and the Singapore cases make them increasingly explicit. For Singapore, the proper response is not to search for a hidden unity in English law or to adopt a general prima facie tort of intentional economic harm. It is to coordinate the torts while preserving the distinctions that do real limiting work. The field’s defensible coherence lies in disciplined differentiation: tort by tort, identify the protected interest, the wrongful interference, the relevant intention and the adjacent legal regime. That method offers a more candid and stable basis for future development than monism.

246

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

THE QUISTCLOSE TRAP: INTENTION IN QUISTCLOSE TRUSTS Navin Sivakumar* Winner of Best Article Award 2026 ABSTRACT The article defends the test for the Quistclose trust proposed by Lord Millett in Twinsectra Ltd v Yardley [2002] 2 AC 164, namely that a Quistclose trust arises when the parties did not intend the property transferred to be at the free disposal of the transferee. It rejects flawed alternatives that have been proposed in the case law in both the UK and Singapore. The test of free disposal is also normatively compelling. It captures the unconscionability tackled by Quistclose trusts. This is, helpfully, the same kind of unconscionability that justifies resulting trusts. I.

INTRODUCTION

The Quistclose trust is a prime example of how deeply equity has seeped into modern commerce, but also how easily equity might lose its way.1 Since it was first articulated by the House of Lords in Barclays Bank Ltd v Quistclose Investments Ltd [1970] AC 567 (“Quistclose”), the Quistclose trust has been accepted and reaffirmed by various common law jurisdictions, but has proved difficult to integrate into the orthodox taxonomy of trusts. 2 The nature and scope of the Quistclose trust remain uncertain. For those who have transferred assets for a specific purpose and now seek recovery, the Quistclose trust is a fickle remedy. 3 But these problems can be cured. It is possible to discern a category of situations in which a Quistclose trust can justifiably arise, as the House of Lords did in Quistclose and Twinsectra Ltd v Yardley [2002] 2 AC 164 (“Twinsectra”).4 It is true that one cannot look to the standard taxonomy of express, resulting or constructive trusts to identify a test for the Quistclose trust. However, it only muddies the waters to say that a Quistclose trust might “be either express or resulting”, as the Singaporean High Court did in Attorney-General v Aljunied-Hougang-

Hertford College, University of Oxford, BA Jurisprudence, Class of 2027 Lord Millett, ‘Foreword’ in William Swadling (ed), The Quistclose Trust, Critical Essays (Hart Publishing 2004) 2 Barclays Bank v Quistclose Investments Ltd [1970] AC 567 3 Ewan Mckendrick, ‘Commerce’ in William Swadling (ed), The Quistclose Trust, Critical Essays (Hart Publishing 2004) 4 Twinsectra Ltd v Yardley [2002] 2 AC 164 1

247

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

Punggol East Town Council [2015] SGHC 137 (“AHPETC”).5 If the taxonomy does not offer a self-evident answer, the response should be to look beyond rigid taxonomies to more basic non-taxonomical principles. This article argues that Lord Millett’s formulation in Twinsectra—that a Quistclose trust arises when the parties did not intend for the assets transferred to be at the “free disposal” of the recipient - is the most cogent test for the Quistclose trust, and should be followed. 6 A survey of rival formulations in both English and Singaporean jurisprudence reveals that they suffer from difficulties of their own, and do not improve upon Lord Millett’s analysis. Lord Millett’s test is also backed by principle. The key to my argument is the much-maligned concept of unconscionability. The idea of unconscionability lies at the bedrock of trusts, but tends to be overlooked. It is often believed that the tripartite taxonomy of trusts is sufficiently helpful to understand the law. But express, resulting and constructive trusts are simply ways in which equity responds to different kinds of unconscionability. I will examine the kinds of unconscionability tackled by express and resulting trusts, and show that the Quistclose trust responds to essentially the same species of unconscionability as resulting trusts. This is also most faithfully reflected in Lord Millett’s test, which is why it is also the most principled test and should be used to determine the scope of the Quistclose trust. It is true that powerful policy arguments have been levelled against the Quistclose trust. 7 But questions of policy are beyond the scope of this article. This article has a more modest aim: it simply seeks to set out how the law can be placed on a more principled basis, in terms that are already acceptable to the law. Neither English nor Singaporean courts frame the scope of trusts by reference to freestanding policy justifications, and would likely be wary of doing so. Properly understood, the Quistclose trust is equipped to grapple with the issues of modern commerce for which it was created. It does not need to be relegated to the graveyard of shortlived attempts at the “commercialisation of equity”. 8 Equity’s intervention is justified when the parties intended that the assets transferred would not be at the transferee's free disposal. With

Attorney-General v Aljunied-Hougang-Punggol East Town Council [2015] SGHC 137 [114] Twinsectra Ltd v Yardley [2002] 2 AC 164 [74] 7 Emily Hudson, ‘A Normative Approach to the Quistclose Trust’ (2017) 80 MLR 775 8 Man Yip and James Lee, ‘The Commercialisation of Equity’ (2017) 37 Legal Studies 647 6

248

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

the fundamentals of the Quistclose trust resolved, the more fertile debate for the law would be to focus on the question of “free disposal”. Part II traces the development of the Quistclose trust through the seminal cases of Quistclose and Twinsectra. Part III examines the English and Singaporean jurisprudence on the Quistclose trust, and evaluates the various ways in which courts have formulated the requirements of the Quistclose trust. Part IV shows that it is Lord Millett’s test that best captures the unconscionability involved in a Quistclose trust. Part V considers how the law should instead direct its attention to the more productive question of “free disposal” to achieve even greater clarity on the circumstances in which a Quistclose trust should be found. II.

THE ORIGINS OF THE QUISTCLOSE TRUST

A. The foundations: Barclays Bank v Quistclose Investments The Quistclose trust takes its name from the decision of the House of Lords in Quistclose, arising from a claim by a lender, Quistclose Investments, that the money held in the Barclays Bank account of the now-insolvent borrower, Rolls Razor Ltd, was held on trust for Quistclose. Quistclose and Rolls Razor had agreed that the loan would “only be used to meet [a] dividend” that had been declared by Rolls Razor for its creditors (who were also shareholders). 9 Since the dividend was never paid, Quistclose argued that the beneficial interest remained with them throughout, so Barclays could not set the money off against Rolls Razor’s overdraft on other accounts. Lord Wilberforce (with whom all the other Lords agreed) accepted that a trust in favour of Quistclose had arisen. Based on the terms on which Quistclose had extended the loan to Rolls Razor, he assessed that it was “made only so as to enable Rolls Razor Ltd to pay the dividend and for no other purpose”. 10 This created a primary trust in favour of Rolls Razors’ creditors, but upon the failure of the primary trust, a secondary trust in favour of the settlor, Quistclose. This trust was created by the “mutual intention” between the Rolls Razor and Quistclose that the sum advanced be “used exclusively for payment of a particular class of its creditors”, i.e. those entitled to the dividend, and “should not become part of the assets” of Rolls Razor. 11

Barclays Bank v Quistclose Investments Ltd [1970] AC 567, 569 ibid 580. 11 ibid. 10

249

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

There is some disentangling required to understand Lord Wilberforce’s conclusion. There at least three different limbs to what Lord Wilberforce described as the “mutual intention” of Rolls Razor and Quistclose, which I label the primary intention, the secondary intention and the corollary: 1. The primary intention: that the sum should instead “be used exclusively for payment of a particular class of creditors” 2. The secondary intention: that the sum advanced “should not become part of the assets of Rolls Razor Ltd” 3. The corollary: by “necessary consequence” and by “process simply of interpretation (emphasis added)”, if for any reason, the “dividend could not be paid, the money was to be returned to [Quistclose]”12 At the outset, it should be noted that only the primary intention appears to faithfully reflect the express agreement of Rolls Razor and Quistclose. After all, the agreement was simply that the “amount will only be used to meet the dividend due”.13 The secondary intention identified by Lord Wilberforce, that the money should not form part of Rolls Razor’s assets, is problematic. First, while framed as an “intention”, it does not seem to describe as a matter of fact what the parties intended, but rather the legal effect of the agreement that they had entered into. Second, as Lord Millett asked in Twinsectra, if the parties had supposedly agreed that Rolls Razor would not beneficially own the sum, “where in such a case is the beneficial interest pending application of money for the stated purpose”? 14 Lord Wilberforce appears to suggest that the “primary trust” was in favour of Rolls Razor’s creditors, who were the intended beneficiaries of the loan. But this is not tenable. On this view, the money was beneficially owned by the creditors even before Rolls Razor paid out the dividend, while it remained in Rolls Razor’s bank account. This could not have been what was intended. While the dividend was meant to reach the creditors, it is absurd to say that the settlor and Rolls Razor agreed to generate equitable rights for the creditors over a particular sum of money while still in Rolls Razor’s bank account, creating a relationship of trusteeship between Rolls Razor and the creditors. So it is difficult to accept Lord Wilberforce’s assertion that the parties intended that the sum

ibid. ibid 579. 14 Twinsectra Ltd v Yardley [2002] 2 AC 164 [79] 13

250

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

advanced should not form part of Rolls Razor’s assets, especially when he identifies the true beneficial owner as the creditors. Likewise, Lord Wilberforce’s corollary, that the parties intended that the money should be returned to the settlor when the dividend could not be paid is not a true factual inference. Swadling rightly argues that there was no proof of such intention. 15 In which case, Lord Wilberforce might have been making a disguised argument for an automatic resulting trust. An automatic resulting trust arises where the declared trust fails, or has failed to exhaust the whole beneficial interest, as took place in Vandervell v IRC [1967] 2 AC 291.16 But this does not work either. Swadling has also shown that where trusts for persons are concerned, here, a trust for the creditors entitled to the dividend, there can be no subsequent failure. The creditors either acquired the beneficial interest in the funds when the funds were transferred to the bank, or never acquired a beneficial interest at all. There is no scope for subsequent failure. The trust either fails at the outset (i.e. at the point of transfer) or does not fail at all. The trust that arose in Quistclose thus cannot be explained as an automatic resulting trust, nor as part of a series of successive trusts. 17 Thus, while Quistclose identified the basic trigger of the Quistclose trust: an intention that the sum transferred should be used exclusively for a particular purpose, it never convincingly explained the steps by which this intention created a trust. B. The modern orthodoxy: Twinsectra v Yardley In any event, while later cases (and other jurisdictions) have accepted that the facts in Quistclose gave rise to a trust, Lord Wilberforce’s approach has not been followed. His reasoning was dismantled by Lord Millett in Twinsectra, whose analysis now forms the orthodoxy on the Quistclose trust.18 For Lord Millett, the fatal flaw in Lord Wilberforce’s approach was his view that a primary trust had been created in favour of the creditors, the intended beneficiaries of the loan. This unduly narrowed the scope of the Quistclose trust, since some loans might have no identifiable

William Swadling, ‘Orthodoxy’ in William Swadling (ed), The Quistclose Trust, Critical Essays (Hart Publishing 2004) 16 Vandervell v IRC [1967] 2 AC 291 17 William Swadling, ‘Orthodoxy’ in William Swadling (ed), The Quistclose Trust: Critical Essays (Hart Publishing 2004) 18 Pricky Bay Waterside v British American Insurance Co Ltd [2022] UKPC 8 [29]

251

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

beneficiaries, as was the case in Twinsectra itself. They may be made for “an abstract purpose with no one but the lender to enforce performance or restrain misapplication of the money”, and there is “no reason to make an arbitrary distinction between money paid for an abstract purpose and money paid for a purpose which can be said to benefit an ascertained class of beneficiaries”.19 If it is the agreement between the settlor and the borrower that impresses the loan with a trust, it is not clear why it should matter whether the purpose of the trust is abstract or for particular persons. In Twinsectra itself, the purpose of the loan was to finance the acquisition of property. This was the only purpose to which the lender had agreed to apply the money, and the lender sought a personal undertaking from the solicitor that the money would be retained by the solicitor “until they were applied in the acquisition of property” and that it would be “utilised solely for the acquisition of the property on behalf of [the] client and for no other purpose”. 20 This was enough for the House of Lords to conclude that a Quistclose trust arose when the funds were received by the solicitor. This was because, in Lord Millett’s view, it was “well established that a loan to a borrower for a specific purpose where the borrower is not free to apply the money for any other purpose created fiduciary obligations on the part of the borrower which a court of equity will enforce”.21 Until the money was applied to the designated purpose, it would be held in resulting trust for the lender, who retained a beneficial interest. For such a trust to arise, parties must not have intended for “the money to be at the free disposal of the recipient”, based on the “intention of the parties collected from the terms of the arrangement and the circumstances of the case”. 22 Perhaps most helpfully, Lord Millett set out a justification of the Quistclose trust, which went unaddressed in Quistclose itself. The imposition of a trust was justified because it was “unconscionable for a man to obtain money on terms as to its application and then disregard the terms on which he received it”. This duty was “not contractual, but fiduciary”, because “a person who makes money available on terms that it is to be used for a particular purpose only and not for any other purpose thereby places his trust and confidence in the recipient to ensure that it is properly applied”. This was a “classic situation in which a fiduciary relationship

Twinsectra Ltd v Yardley [2002] 2 AC 164 [89] ibid [9]. 21 ibid [68]. 22 ibid [69], [74]. 20

252

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

arises, and since it arises in respect of a specific fund it gives rise to a trust (emphasis added)”. 23 The recipient receives the money only to apply it to the designated purpose, which is why the lender retains the beneficial interest even after the transfer. The recipient simply takes custody of the money to carry out the shared purpose. Given Lord Millett’s rather far-reaching justification for the Quistclose trust, he found that it arose in Twinsectra despite two potentially material differences with Quistclose: 1. The transferee and transferor in Twinsectra were not borrower and lender, as in Quistclose. The transferee in Twinsectra was a solicitor who acted as an intermediary for the money to reach another. 2. The sum advanced could be for an abstract purpose, such as the acquisition of property. No identifiable beneficiary was required, as noted above. The final point that emerges from Lord Millett’s analysis in Twinsectra is the juridical basis of the Quistclose trust. Lord Millett determined that the Quistclose trust was an “entirely orthodox example of the kind of trust known as a resulting trust”.24 The legal title is transferred, but the recipient has no “beneficial interest in the money”, simply a limited power to apply it “for the stated purpose or return it”. Should the purpose fail, the money is returnable to the lender because the resulting trust is “no longer subject to any power” for the recipient to make use of the money.25 In his analysis, Lord Millett appears to endorse Chambers’ thesis on resulting trusts: resulting trusts arise when there is a “transfer of property in circumstances in which the transferor did not intend to benefit the recipient” and they respond to the “absence of the intention on the part of the transferor to pass the entire beneficial interest, rather than to any positive intention to retain it”. 26 It should be noted, however, that this view that resulting trusts arise in response to an absence of intention to benefit the recipient was expressly rejected by Lord Browne-Wilkinson in Westdeutsche Landesbank Girozentrale v Islington London Borough Council [1996] AC 669 (“Westdeutsche”).27 That being said, Westdeutsche rejected the lack-of-intention view in the context of a mistaken payment claim and Birks’ wider thesis on how resulting trusts responded 23

ibid [76]. ibid [100]. 25 ibid. 26 ibid [92]. 27 Westdeutsche Landesbank Girozentrale v Islington London Borough Council [1996] AC 669 24

253

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

to unjust enrichment. 28 As I will explain below, it might be possible to take a somewhat narrower view of what the ‘lack of intention’ to benefit the transferee involves so as to remain within the bounds of trust orthodoxy. Taken together, in Twinsectra, Lord Millett might not have made a completely successful case for the juridical classification of the Quistclose trust as a resulting trust, and so the academic controversy has persisted. But his true accomplishment was in setting out the ingredients of the Quistclose trust, and as I will explain, identifying a test for the Quistclose trust that captured the normativity, namely the unconscionability, involved. That is, the parties’ intention that the assets not be at the free disposal of the transferee. C. After Twinsectra There are obvious advantages to Lord Millett’s formulation, at least when juxtaposed against Lord Wilberforce’s alternative. For a court to ask whether the parties intended the money to be at the free disposal of the recipient, as Lord Millett proposed, is a much stronger indication of whether the party should be beneficially entitled than whether they intended the sum should “be used exclusively for payment of a particular class of [persons]”, as suggested by Lord Wilberforce.29 This is mainly because Lord Wilberforce’s test is over-inclusive. Financial intermediaries, such as banks, exchanges or investment schemes, often receive funds specifically for the benefit of a particular person, whether the settlor or a third party. Yet, it is well-established that these intermediaries acquire both the legal and beneficial interest in the funds transferred to them.30 Lord Wilberforce’s test focuses on the motive of the transfer at the expense of the manner in which the parties intended for the transferee to hold the funds. D. The position in the UK and Singapore Given the strength of Lord Millett’s analysis, it is unsurprising that Twinsectra has been accepted as the leading authority on Quistclose trusts in the UK. Twinsectra was accepted as the “core analysis” in Pricky Bay Waterside v British American Insurance Co Ltd [2022] UKPC 8 (“Prickly Bay”), as the “now accepted analysis” in Ali v Dinc [2020] EWHC 3055 (“Ali v

ibid 709 - 715. Barclays Bank v Quistclose Investments Ltd [1970] AC 567, 580 30 First City Monument Bank plc v Zumax Nigeria Ltd [2019] EWCA Civ 294 29

254

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

Dinc”) and cited with approval in a number of other cases.31 The Quistclose trust was also accepted as part of Singapore law in the High Court case of Pacific Rim Palm Oil Ltd v PT Asiatic Persada [2003] SGHC 243 (“Pacific Rim”). 32 The Quistclose trust is now governed by the principles set out by Quentin Loh J in AHPETC, which took the view that “Lord Millett’s model in Twinsectra should be adopted as the law in Singapore”. 33 AHPETC was endorsed and applied by Andrew Ang SJ in MSP4GE Asia Pte Ltd and another v MSP Global Pte Ltd and others [2019] 3 SLR 1348 (“MSP4GE”), and by Goh Yihan J in Envy Asset Management v CH Biovest [2024] SGHC 46 (“Envy”).34 However, courts in both jurisdictions have reformulated Lord Millett’s test in various ways. The alternative formulations in the case law now include (1) an intention to create a trust (Bieber and others v Teathers Ltd [2012] EWCA Civ 1466 (“Bieber”), (2) an intention to retain a beneficial interest (Prickly Bay, AHPETC) and (3) that the transferee has the “necessary proprietary understanding” (Ali v Dinc). All these proposed alternatives are untenable (Section II). In addition, they fail to capture the unconscionability involved in the Quistclose trust, which Lord Millett’s test accomplishes (Section III). Another development in both jurisdictions is also worth brief comment, since it bears on the analysis below. In both the UK and Singapore, it has been suggested by some judges that, depending on the circumstances, the Quistclose trust may be either express or resulting. 35 There is very little to commend this view, as Andrew Ang SJ observed in MSP4GE Asia. 36 It does not resolve the difficulties involved in the resulting Quistclose trust, which forms the focus of this article. It obscures the legal and normative issues involved in the resulting Quistclose trust, and encourages assimilation to an express trust. And it appears entirely superfluous. There are well-worn legal techniques to establish whether an express trust has been declared, and there appears to be no clear benefit to saying that some of these express trusts are now additionally “Quistclose trusts”.

Pricky Bay Waterside v British American Insurance Co Ltd [2022] UKPC 8 [29]; Ali v Dinc [2020] EWHC 3055 [230]; Bellis v Challinor [2015] EWCA Civ 59 [54] - [66]; Bieber and others v Teathers Ltd [2012] EWCA Civ 1466 [14] - [15] 32 Pacific Rim Palm Oil Ltd v PT Asiatic Persada [2003] SGHC 243 33 Attorney-General v Aljunied-Hougang-Punggol East Town Council [2015] SGHC 137 [110] 34 MSP4GE Asia Pte Ltd and another v MSP Global Pte Ltd and others [2019] 3 SLR 1348; Envy Asset Management v CH Biovest [2024] SGHC 46 35

Attorney-General v Aljunied-Hougang-Punggol East Town Council [2015] SGHC 137 [110]; Pricky Bay Waterside v British American Insurance Co Ltd [2022] UKPC 8 [32] 36 MSP4GE Asia Pte Ltd and another v MSP Global Pte Ltd and others [2019] 3 SLR 1348, [110]

255

Singapore Comparative Law Review

III.

The Quistclose Trap: Intention In Quistclose Trusts

EVALUATING THE ALTERNATIVES A. Intention to create a trust

On occasion, the Quistclose trust is treated as an express trust in all but name. This happened in the English Court of Appeal case of Bieber, where Patten LJ added a gloss to the Twinsectra test. It was not enough that “the money when paid was not at the free disposal of the payee”. 37 In addition, “the contractual or other arrangements [must have been] intended to provide for the preservation of the payor’s rights and the control of the use of the money through the medium of a trust (emphasis added)”.38 This is irreconcilable with both Quistclose and Twinsectra. Neither case would have surmounted the formidable threshold that Patten LJ describes, because neither transferor envisioned the continued control of assets transferred through a trust. This also renders the Quistclose trust superfluous, because if the contractual arrangements indicate that the use of the funds transferred would be subject to a trust, then an express trust has been created. But the Quistclose trust was introduced precisely to resolve situations where parties had not agreed, whether expressly or impliedly, to set up a trust. The transferee is deemed to hold for the benefit of the transferor, subject to the transferee’s power to apply the funds to the shared purpose. That is the substance of the Quistclose trust. Likewise, in Bellis v Challinor [2015] EWCA Civ 59, while Briggs LJ (as he then was) faithfully applied the test of free disposal, he commented in obiter that even in a Quistclose trust, ““[t]here must be some intention to create a trust on the part of the transferor”. He added that this meant that “the transferor must have intended to enter into arrangements which, viewed objectively, have the effect in law of creating a trust (emphasis added)”.39 The latter half of Briggs LJ’s statement is unobjectionable: it is uncontentious that the subjective intentions of the parties involved are irrelevant. But to equate an intention to enter into arrangements which objectively create a trust with “some intention to create a trust” is difficult to accept. The two intentions should not be conflated. Only in express trusts do parties intend to create a trust. In non-consensual trusts, such as resulting and constructive trusts, it is some other intention, agreement or conduct which produces the trust. In those situations, the requisite intention falls short of an intention to create a trust. For instance, as far as Singapore law is concerned, a presumed resulting trust where there is lack of intention to benefit the transferee.40

Bieber and others v Teathers Ltd [2012] EWCA Civ 1466 [15] ibid. 39 Bellis v Challinor [2015] EWCA Civ 59, [57] 40 Chan Yuen Lan v See Fong Mun [2014] SGCA 36 [44] 38

256

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

There is some debate in English law as to whether it is the lack of intention to benefit the transferee or the intention to keep a beneficial interest which creates a presumed resulting trust, but either way, where a resulting trust is concerned, the parties involved have not intended, expressly or impliedly, to create a trust. Briggs LJ’s remark thus risks confusion and might be seen to assimilate Quistclose trusts to express trusts. A slightly different error was made by Belinda Ang J in Pacific Rim, which appears to be the first case in Singapore where a Quistclose trust was successfully claimed. 41 The plaintiffs had advanced funds to the defendant to repay existing debts, but those debts were thereafter auctioned off and became incapable of settlement, so the plaintiffs sought recovery of the funds advanced. In many respects, this was an orthodox Quistclose case, given the agreement between the parties that the money would be used to settle a particular debt with a certain entity. But Ang J erred when she took the fact that loan payments were made into a separate account as a “factor tending to suggest that a Quistclose trust was intended”. 42 This is a peculiar statement. It is one thing to infer from the facts that the parties intended that the transferee would hold certain assets on trust for the transferor, but it is another to suggest that the parties intended that a Quistclose trust would arise. Especially since Ang J had earlier accepted that the Quistclose trust was a resulting trust, the claim that the parties intended to create a resulting trust is a contradiction in terms. 43 As stressed above, if a trust arose because the parties so intended, then an express, not a resulting, trust has been created. In addition, it is generally not the case that parties must have intended the legal consequences of a particular arrangement; all that is required for those consequences to apply is for parties to have intended the arrangement itself. And if the question was truly whether a Quistclose trust was intended, the analysis in the case would have looked quite different. It would not have focused quite so heavily on the restrictions on the use of the transferred money, but whether the parties had made an adequate declaration of a Quistclose trust through those arrangements. Ang J’s remark cannot be reconciled with her wider analysis she undertook in Pacific Rim when evaluating whether a Quistclose trust arose. Thus, while Pacific Rim was a classic example of a Quistclose trust, it was not entirely without issue on the prerequisites for a Quistclose trust.

Pacific Rim Palm Oil Ltd v PT Asiatic Persada [2003] SGHC 243 ibid [26]. 43 ibid [19]. 42

257

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

B. An intention to retain some beneficial interest An alternative test is that of the Privy Council in Prickly Bay.44 As with other authorities, it endorsed Lord Millett’s speech in Twinsectra, yet it took a position at odds with Twinsectra on the intention required to constitute a Quistclose trust.45 Given that it was relatively straightforward to establish that the sum advanced was intended to be at the free disposal of the transferee, Prickly Bay’s variation of the Twinsectra test is difficult to understand. The dispute in Prickly Bay concerned a sum paid by the claimant to an insurance company for an annuity plan, intended to cover a liability due in two years’ time. In turn, the insurer guaranteed that the payout would be made. Naturally, the Eastern Caribbean Court of Appeal found that there was no understanding that the money deposited should not be at the free disposal of the insurer, which should have been sufficient to dispose of the matter. 46 But the approach taken by the Privy Council was more circuitous. Lady Arden took the view that the true intention involved in constituting a Quistclose trust, even as a resulting trust, was an intention that the “payer should retain some beneficial interest in the funds”.47 There are two issues with Lady Arden’s test. First, to ask whether the parties intended that the payer would retain some “beneficial interest” is simply to beg the question. As explained, the Quistclose trust is only invoked where parties themselves have not made any agreement as to the fate of the beneficial interest, but have, for instance, agreed that the transferee’s use of the funds should be restricted to a specific purpose. While the effect of a Quistclose trust is that the beneficial interest remains with the transferor, it is quite another matter to say that parties intended that this should be the case. The second difficulty with Lady Arden’s test is that it neglects how, as Lady Arden herself stressed, the Quistclose trust is a “default trust”. It will not “override any arrangement which the parties may have agreed for the destination for the funds previously earmarked for the purpose”. 48 In other words, if the parties had agreed on the location of the beneficial interest, whether in the transferor or the transferee, there would be no scope for the Quistclose trust to

Pricky Bay Waterside v British American Insurance Co Ltd [2022] UKPC 8 ibid [29]. 46 ibid [2] - [4]. 47 ibid [31]. 48 ibid [25]. 45

258

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

arise. Lady Arden’s reasoning is thus self-contradictory: the Quistclose trust cannot be both a “default trust” and arise in response to a shared intention as to the “beneficial interest”. Loh J’s test in AHPETC is similar, but sidesteps some of these difficulties. In AHPETC, the Ministry of National Development sought a declaration that it had a beneficial interest in the grants made to a particular town council, such that the town council was bound to apply the funds for particular purposes. Loh J rejected that a Quistclose trust had arisen, and held that the intention required for a “resulting Quistclose trust is not an extant intention, but an absence of intention to pass the entire beneficial interest to the recipient”. 49 A “finding that there was an intention to pass the entire beneficial interest would therefore foreclose a resulting Quistclose trust from arising”.50 This aligns the test for Quistclose trusts with that of resulting trusts, which in the Singaporean Court of Appeal case of Chan Yuen Lan v See Fong Mun [2014] SGCA 36 (“Chan Yuen Lan”), was accepted to arise in response to a lack of intention to benefit the transferee.51 But Loh J’s test is not without issues of its own, which are mostly practical rather than conceptual. As Loh J observes, while the test looks to the “absence of intention”, in practice it would be a positive intention that would establish a Quistclose trust, such as an intention to pass the entire beneficial interest to the transferee that would negative a Quistclose trust, or an intention to retain a beneficial interest (as Lady Arden proposed). 52 In effect, such a test asks whether parties had agreed on the location of a beneficial interest, which is by definition absent in Quistclose trusts, because the Quistclose trust is a default trust, as explained above. A viable test for the Quistclose trust thus cannot focus on the effect of the transfer on the beneficial interest, because that is the very question at issue. In many cases it will not be possible to make a factual finding as to what parties intended for the beneficial interest. If parties had agreed on the fate of the beneficial interest, it would be the law on express trusts rather than resulting trusts should apply. C. Necessary proprietary understanding Ali v Dinc offers the most unusual gloss on the Twinsectra principles, albeit in obiter. In Ali v Dinc, the claimant argued that he had transferred properties to the defendant solely to raise funds which would then be transferred to the claimant. The defendant failed to raise the funds, and the claimant sought to recover the properties. It was found that a Quistclose trust arose, 49

Attorney-General v Aljunied-Hougang-Punggol East Town Council [2015] SGHC 137 [123] ibid. 51 Chan Yuen Lan v See Fong Mun [2014] SGCA 36 [44] 52 Attorney-General v Aljunied-Hougang-Punggol East Town Council [2015] SGHC 137 [123] 50

259

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

because the transferee had the “necessary proprietary understanding”. 53 The judge, Sarah Worthington KC (Hon) sitting as Deputy High Court Judge, endorsed Twinsectra but then added that where the transferee was concerned, it was “absolutely crucial that the recipient of the property understands that his use of the property is specifically restricted: he is not entitled to use the property as his own despite his legal ownership of it. Of course it does not matter whether [the recipient] understands the legal niceties - it is rare that he will - but it is necessary that he understands that his newly acquired legal ownership does not come with all the normal incidents of full ownership. That is the case here, on [the defendant’s] own evidence”. 54 The judge’s approach to the necessary intention for the transferor was more orthodox: as the transferor was concerned, what was important was that “the transferor of the property intends the type of arrangement which a court holds gives rise to such a trust, but there is really no need for [the transferor] to understand that this has any proprietary consequences”. 55 The judge’s approach towards the transferee is misconceived. As Lord Millett took pains to stress in Twinsectra, the subjective intentions of the parties who enter into an agreement that gives rise to a Quistclose trust are “irrelevant”. If the transferee “enters into arrangements which have the effect of creating a trust, it is not necessary that he should appreciate that they do so; it is sufficient that he intends to enter into them”. 56 Nor are the subjective intentions of the transferee “relevant to the construction of the [transferee’s] undertaking”. 57 The reason for this is self-evident: there must be certainty in how trusts arise. Moreover, if the judge’s view is taken to its logical conclusion, the Quistclose trust arises not upon receipt by the transferee, but when the transferee acquires subjective knowledge of the extent to which use of the assets is restricted. The judge thus seems to be mistakenly transplanting the logic of constructive trusts to Quistclose trusts. It is true that a constructive trust, for instance in relation to receipt of mistaken payments, springs into being at the moment of unconscionable knowledge, not at the moment of receipt.58 But this is because mistaken payments are beneficial transfers to begin with, where something more is required for equity to intervene. By contrast, the whole point with the Quistclose trust is to identify and respond to non-beneficial transfers. The obiter in Ali v Dinc thus sits uneasily with the academic and judicial consensus that where Quistclose trusts

Ali v Dinc [2020] EWHC 3055 [240] ibid [245]. 55 ibid. 56 Twinsectra Ltd v Yardley [2002] 2 AC 164 [71] 57 Twinsectra Ltd v Yardley [2002] 2 AC 164 [72] 58 Westdeutsche Landesbank Girozentrale v Islington London Borough Council [1996] AC 669, 705 54

260

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

are concerned, it is not the recipient’s subjective knowledge, but the transfer that is trustgenerating. This survey of post-Twinsectra authorities reveals that courts in both jurisdictions remain conflicted on the intention required to create a Quistclose trust. Although Twinsectra was cleareyed on the law and has been endorsed in both jurisdictions, in practice, courts have imposed somewhat stricter requirements before a Quistclose trust is found. As shown, these alternatives do not withstand logical scrutiny. The difficulty often arises from how courts, on the one hand, accept that the Quistclose trust is a “default” trust or a resulting trust, and yet require there to have been an agreement between the parties as to the beneficial interest. Once it is either expressly or impliedly agreed that the assets are to be held for the benefit of the transferor, an express trust is created. The Quistclose trust becomes wholly superfluous. IV.

UNCONSCIONABILITY AND THE QUISTCLOSE TRUST

Quite apart from the analytical difficulties with the alternatives, there are powerful arguments of principle in favour of using Lord Millett’s test of free disposal to determine the scope of the Quistclose trust. It captures the unconscionability involved in the Quistclose trust. It has become necessary to turn to unconscionability because the intention that produces a Quistclose trust cannot be derived from the taxonomy. This much has been accepted by courts, who have taken it upon themselves to formulate tests of their own. Even where courts have accepted that the Quistclose trust is a resulting trust, neither the test for presumed nor automatic resulting trusts has been applied. It would therefore be entirely sterile to continue debating how the Quistclose trust should be categorised. Rather, the labels should be pierced to reveal the deeper principle involved, namely, unconscionability. The unconscionability that animates the Quistclose trust was explained by Lord Millett in Twinsectra. A Quistclose trust arises because it is “unconscionable for a man to obtain money on terms as to its application and then disregard the terms on which he received it.” A person who “makes money available on terms that it is to be used for a particular purpose only and not for any other purpose places his trust and confidence in the recipient”, this being a classic situation in which a “fiduciary relationship arises”. He adds that since the fiduciary relationship arises “in respect of a specific fund it gives rise to a trust”.59

Twinsectra Ltd v Yardley [2002] 2 AC 164 [76]

261

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

At the outset, it might be objected that it is far too arbitrary to impose a trust on the grounds of “unconscionability”, especially if it emanates from simple “trust and confidence”. To Hudson, for instance, Lord Millett failed to explain why the trust and confidence in the Quistclose trust was any different from promises in the law of contract. 60 Virgo also takes issue with the term “unconscionability”, which conceals a “myriad of meanings”. 61 In any event, none of these objections are germane to my argument. This is because it is not necessary to rely on some abstract unconscionability to determine the scope of the Quistclose trust. Rather, I will examine the unconscionability involved in well-established trusts to show that the Quistclose trust can be rationalised in the same way as existing trusts, if subject to Lord Millett’s test of free disposal. This would be a powerful reason to prefer Lord Millett’s test. There is also a compelling case that unconscionability is not just the normative basis, but the legal principle behind the beneficiary’s right. At first glance, it might appear difficult to see how unconscionability could possibly justify the beneficiary’s right priority over an innocent creditor. It might, of course, be answered that the debtor’s assets held on trust for the beneficiary were never part of the debtor’s assets, and thus were never available to the creditor. But this is a circular argument, because it presupposes the nature of the beneficiary’s right. The true answer is that while the creation of any given beneficial right is grounded in unconscionability, this does not necessarily apply to how beneficial rights are enforced. The beneficiary’s right, at its inception, was not enforceable against the creditor. The sequence of steps by which the beneficiary’s rights came to be enforced against an ever-expanding class of individuals was masterfully traced by Maitland. 62 At first, the beneficiary only had a right against the trustee, who had undertaken to hold the land for him. The next step was to enforce the trust against those who succeeded to the trustee’s title, because they sustain “wholly or partly the persona” of the original trustee. Enforcement gradually enlarged to creditors, then donees, purchasers with actual knowledge, and finally purchasers with constructive knowledge, stopping short of the bona fide purchaser for value. The essential point, therefore, is that it is important to distinguish how the beneficial right comes into existence from how it has come to be enforced. How trusts have come to be enforced will not explain the way in which beneficial interests are created in the first place. To understand how beneficial interests 60

Emily Hudson, ‘A Normative Approach to the Quistclose Trust’ (2017) 80 MLR 775 Graham Virgo, ‘Whose Conscience? Unconscionability in the Common Law of Obligations’ in A Robertson and M Tilbury (eds), Divergences in Private Law (Oxford: Hart, 2016) 62 FW Maitland, Equity: A Course of Lectures (AH Chaytor and WJ Whittaker eds, 2nd edn, Cambridge University Press 1936) 61

262

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

arise, including in the Quistclose trust, it is instead necessary to examine the various triggers for a trust, and the unconscionability involved in these triggers. A. First principles in trusts Express trusts Most commonly, a trust arises in response to a declaration of trust over a sufficiently certain subject-matter, in favour of a sufficiently certain object. But it is not necessary for the settlor to have intended the various consequences that a trust might create. As Briggs LJ observed in Bellis v Challinor all that matters is for the settlor, or transferor, to have “intend[ed] the type of arrangement which a court holds gives rise to a trust”. There is “really no need” to understand the “proprietary consequences” involved. 63 It is not even necessary for parties to know that the effect of the declaration is to create what is technically known as a “trust”, because the declaration is to be construed objectively not subjectively. As Snell’s Equity notes, the court looks to the “substance and effect of the words used, against the background of any relevant surrounding circumstances”. 64 As such, in Paul v Constance [1977] 1 WLR 527 (“Paul v Constance”), it was held that an express trust arose where the defendant had told the claimant that money in a shared account used for bingo games was “as much yours as mine”. 65 But the principle of objectivity alone cannot explain why the declaration in Paul v Constance could create a trust. An objective approach merely determines how words or conduct are to be construed, not the standard to which they are then held. That instead turns on what is contained in an objective intention to create a “trust”. Agnew and Douglas have produced a credible answer. 66 They have identified two separate intentions that must be manifested to declare a trust. First, an intention to “hold rights for the benefit of another”. Second, an intention “that by the settlor’s very words or action, the trust relationship should be immediately constituted”. The former is particularly revealing. Drawing on Raz, they argue that the undertaking here is far stronger than a contractual promise, because it is a commitment not only to be “partial to the beneficiary, but also to hold her property for the beneficiary and subjugate her own personal interests to those of the beneficiary, where they 63

Bellis v Challinor [2015] EWCA Civ 59, [57] J McGhee and S Elliott (eds), Snell's Equity (35th edn, Sweet & Maxwell 2025), 711; Compania De Navegacion Palomar, S.A. and others v Ernest Ferdinand Perez De La Sala and another matter [2017] SGHC 14 [124] - [126] 65 Paul v Constance [1977] 1 WLR 527 66 S Agnew and S Douglas, ‘Self-declarations of trust’ (2019) 135 LQR 67 64

263

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

may conflict”. It is a pledge to maintain single-minded loyalty to the beneficiary in relation to the trust property.67 The “proprietary consequences” and duties of trusteeship do not have to be specifically intended, either subjectively or objectively, because they are wholly derivative of an undertaking to hold property for the benefit of another. Resulting trusts But a trust can also arise in response to intentions that fall short of an intention to create a trust, as in the case of resulting trusts. The orthodoxy is that there are two types of resulting trusts: automatic resulting trusts and presumed resulting trusts. 68 A presumed resulting trust arises when A has made a gratuitous transfer of property to B, or if A has contributed to the purchase price of property to which B now holds legal title. B will be taken to hold on trust for A, unless B can rebut a presumption that A did not intend to benefit B (per Air Jamaica v Charlton [1999] 1 WLR 1399, and per Chan Yuen Lan in Singapore) or that A intended to keep a beneficial interest (per Westdeutsche).69 It should be briefly noted that the content of the presumption remains contested in English law: while Lord Millett has expressed sympathy for the view that the resulting trust arises in response to the lack of intention to benefit the transferee, this was rejected by Lord Browne-Wilkinson in Westdeutsche. Lord Browne-Wilkinson instead took the view that a presumed resulting trust arose because of the transferor’s intention to keep a beneficial interest, which was why there would be no resulting trusts over mistaken payments, since mistaken payments were beneficial transfers. The position in Singapore law is clearer: the basis of the resulting trust is the lack of intention to benefit the transferee. One advantage of the lack-of-intention view over the positive intention view is that it can also explain automatic resulting trusts. Automatic resulting trusts arise when there has been a transfer on trust, but the trust has failed for any number of reasons, such as want of objects (Vandervell v IRC [1967] 2 AC 291) or for offending the rule against private purpose trusts (Re Endacott [1960] Ch 232).70 The beneficial interest then vests in the settlor, instead of the intended object.

Hopcraft & Anor v Close Brothers Limited [2025] UKSC 33, [7] Re Vandervell’s Trust (No 2) [1974] Ch 269, 294 69 Chan Yuen Lan v See Fong Mun [2014] SGCA 36 [44]; Air Jamaica v Charlton [1999] 1 WLR 1399; Westdeutsche Landesbank Girozentrale v Islington London Borough Council [1996] AC 669, 706 - 715 70 Vandervell v IRC [1967] 2 AC 291; Re Endacott [1960] Ch 232 68

264

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

For the purposes of this section, it is not necessary to resolve the debate over the correct basis of the resulting trust, except to say that every instance of an intention to retain a beneficial interest necessarily involves a lack of intention to benefit the transferee, but that the reverse does not hold true. And as Maitland explains, if the lack-of-intention view is accepted, then the only difference between the presumed and automatic resulting trust is whether this lack of intention is expressed. 71 In a presumed resulting trust, the lack of intention to benefit is unexpressed, which is why a presumption needs to be invoked. By contrast, in an automatic resulting trust, the intention that the transferee should not benefit is expressed, because the transfer was made on trust, but simply failed. But what then is the unconscionability involved in resulting trusts and how does the Quistclose trust fit? B. Unconscionability in trusts While the unconscionability in express trusts arises from the transferee having undertaken to hold property for the benefit of the beneficiary, the unconscionability in resulting trusts is quite different. That automatic resulting trusts are in practice failed express trusts should not obscure this point. The difference is that, unlike an express trust, where the trustee has undertaken to act in the interests of the beneficiary that was designated by the settlor, in an automatic resulting trust, the trustee comes under a duty to the settlor himself, rather than the original beneficiary whose interests the trustee had undertaken to advance. Similarly, in presumed resulting trusts, regardless of whether one adopts the lack-of-intention or positive intention view, the trustee did not come under any duty to the beneficiary because he made a commitment to act in the interests of the beneficiary. The Quistclose trust is the same. The transferee has not undertaken to hold the property transferred for the beneficiary’s interests. Instead, the parties intend the transferee to exercise a power over the property in a way that would necessarily extinguish any trusteeship between the transferee and transferor. The point of the transfer is the realisation of a specific purpose rather than to have the transferee act in the interests of the transferor. In this important regard, Quistclose trusts are more similar to resulting trusts rather than express trusts in that they do not arise from the trustee having undertaken to advance the interests of the beneficiary. But quite apart from differences with the unconscionability in the express trust, there is a powerful case that the Quistclose trust involves the same species of unconscionability as other 71

FW Maitland, Equity: A Course of Lectures (AH Chaytor and WJ Whittaker eds, 2nd edn, Cambridge University Press 1936)

265

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

resulting trusts, which Lord Millett’s test of free disposal captures. In the case of both resulting trusts and the Quistclose trust, although the trustees have not undertaken to hold the property for the benefit of the transferor, the trustees accept property from the transferor knowing that he has placed “trust and confidence” in them in how the property should be used. 72 In presumed resulting trusts this trust and confidence arises regardless of whether one takes the positive intention or absence-of-intention view. Both involve a non-beneficial transfer. In automatic resulting trusts, the transferee was meant to take property as trustee for the original beneficiary, and thus necessarily accepts the property in a context of “trust and confidence”. In the Quistclose trust, the transferee accepts the property so as to apply it to the agreed purpose. The essential difference, therefore, between the unconscionability involved in express trusts, as opposed to the Quistclose and resulting trusts, is that in express trusts the trustee has made a commitment in relation to the beneficiary that he comes under a duty to while in resulting and Quistclose trusts the trustee has made no commitment to act in the interests of the beneficiary that he comes under a duty to, but has agreed to take the property in objectively non-beneficial circumstances. As such, the beneficial ownership that arises in favour of the beneficiary in resulting and Quistclose trusts is a legal consequence rather than the substance of the agreement between trustee and beneficiary. In other words, the unconscionability in resulting and Quistclose trusts arises not from how the trustee takes property as the avowed trustee to the beneficiary as with express trusts, but simply from taking as custodian or steward of specific property. The requirement of specific property is crucial. As Lord Millett pointed out, the Quistclose trust only arose in Twinsectra because the trust and confidence existed in relation to a “specific fund”. 73 If the transferor’s faith in the trustee was not tied to specific property, then it would be difficult to see why the transferor should acquire an interest in relation to the property transferred. This is also where the test of free disposal makes all the difference. Free disposal determines whether the “trust and confidence” placed in the transferee existed only in relation to some shared abstract purpose, or in relation to the specific property that had been transferred. If parties intended the property transferred to form part of the transferee’s general assets, then it cannot be said that the transferee has been entrusted with specific property. By contrast, if

72 73

Twinsectra Ltd v Yardley [2002] 2 AC 164 [76] ibid.

266

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

parties intended that the transferee would not have free disposal of the property, then it can be said that trust and confidence has been imposed in relation to specific property. It is also here that the other formulations considered above falter. To ask whether there was an intention to retain a beneficial interest is not grounded in any particular unconscionable conduct. To ask whether the funds were transferred for a specific purpose glosses over the importance of the transferee’s undertaking in relation to the property transferred. Only the free disposal test probes this crucial normative relationship, and examines the circumstances under which the transferee has taken specific property. IV: THE FUTURE OF THE QUISTCLOSE TRUST: FREE DISPOSAL But even if Lord Millett’s test of free disposal comes to be universally accepted, the curtain should not fall on the Quistclose trust. In this final section, I seek to discuss the future development of the Quistclose trust. It is hoped that the debate might shift to more fertile terrain, namely the essential ingredient of Lord Millett’s test: the idea of free disposal. This would also enable the law to more expressly consider the unconscionability involved in the Quistclose trust. While the question of free disposal has attracted some comment in cases, it has yet to emerge front and centre in discussions on the Quistclose trust. But a few important observations can already be drawn. A. Free disposal in transfers for a particular purpose The test of free disposal has sometimes been erroneously substituted by, or equated with, the question of whether assets were transferred for a specific purpose. While a specific purpose is necessary for a Quistclose trust to arise, and for the transferor to reasonably argue that the transferee’s use of the money is restricted, it is not sufficient. It is only circumstantial evidence for whether the parties intended for assets to be at the free disposal of the transferee. Ang J thus fell into error in Pacific Rim when she conflated the questions of free disposal and specific purpose. She identified the starting point as “whether the advance of US$ 5 million to [the transferees] was to be at the [transferee’s] free disposal or was it for a specific purpose”.

267

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

74 This sets up a false dichotomy: funds can be advanced for a specific purpose (as they often

are) and yet it could be intended that they would be at the free disposal of the transferee. This was precisely the situation in First City Monument Bank plc v Zumax Nigeria Ltd (“Zumax”), an English Court of Appeal case.75 In Zumax, the claimant argued that a Quistclose trust had arisen when it made USD transfers to an account held by its bank with a correspondent bank, which would then credit the claimant’s domestic Naira-denominated account. This arrangement was intended to allow the claimant to exchange foreign USD earnings for Naira. The High Court judge found a Quistclose trust had arisen, because the payment instructions said that the transfers made to the bank’s account were for “further credit to [the claimant]”. This “evince[d] a clear intention” that the funds transferred into the bank’s correspondent account should be held “not for its own benefit but for the benefit of [the claimant]”. 76 The Court of Appeal, rightly, disagreed. While the transfers were made for a specific purpose, there was no evidence that the funds should not be at the bank’s free disposal. Mere identification of the ultimate payee could not impress the funds with a trust. Nor was there any attempt to segregate the transfers from the other inflows and outflows in the account, unlike in Quistclose.77 The wider banker-customer relationship within which the transfers in Zumax took place was also fatal to a Quistclose trust. For the court, the principle that “money placed in the custody of a banker is, [for] all intents and purposes, the money of the banker, to do with it as he pleases” was “long-established and fundamental to how banking is carried on”. 78 Nor was there a parallel to Quistclose, even though Quistclose also involved a claim against a bank. As Lewison LJ put it, “in both Quistclose and in Twinsectra the real contest was between the account holders”. 79 Properly understood, the true issue in Quistclose was whether the funds transferred by Quistclose into Rolls Razor’s Barclays account formed part of Rolls Razor’s free estate (and thus available for Barclays’ set-off) or belonged to Quistclose. What was disputed was the beneficial ownership of a particular personal right (chose in action) held by Rolls Razor against Barclays Bank. In other words, the question in Quistclose was whether there was a trust of a chose in action. In contrast, the thrust of the claim in Zumax was that the funds transferred

Pacific Rim Palm Oil Ltd v PT Asiatic Persada [2003] SGHC 243 [19] First City Monument Bank plc v Zumax Nigeria Ltd [2019] EWCA Civ 294 76 ibid [28]. 77 ibid [32] - [35]. 78 ibid [36]. 79 ibid [78]. 75

268

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

by a customer to its bank created a trust rather than a chose in action. The Court of Appeal was thus entirely correct to see no analogy between Quistclose and Zumax. It was perfectly consistent with the existence of a highly specific purpose that the money was still intended to be at the free disposal of the transferee. A specific purpose is a necessary, but not a sufficient factor for a Quistclose trust to arise, which Lord Millett made clear in Twinsectra. B. Segregation The lack of intention that the funds transferred should be segregated is indicative of free disposal and will often be fatal to a Quistclose trust. As Loh J explained in AHPETC, “at the bottom line, money is at free disposal of the recipient if it falls within the recipient’s general funds; and if the money is to be applied for a special designated purpose it cannot fall within the general funds”. 80 Thus, in AHPETC, where grants made by the Ministry of National Development to a town council were not segregated in a special account and instead deposited into a general account, they were found to be at the free disposal of the transferee and no Quistclose trust arose. As Loh J correctly noted, “purpose cannot be looked at in isolation from free disposal”.81 It is highly unlikely that the fact that the transferee’s wider use of monies is restricted will be sufficient to support a Quistclose trust. This was also at issue in AHPETC: it was argued that a Quistclose trust arose because the town council was “constrained to spend its general funds in accordance with the provisions in [legislation]”. 82 This was rightly rejected. The funds transferred by the Ministry to the town council were not subject to restrictions any stricter than those which applied to the council’s general funds. The monies were thus at the free disposal of the town council, as far as exercising its lawful powers was concerned. Loh J was thus correct to conclude that no Quistclose trust arose. C. Specificity of purpose As noted above, a specific purpose may be powerful evidence of whether the transferee has free disposal, but it is not dispositive. In Envy, Goh J placed weight on the vagueness of the purported purpose of the transfer in concluding that no Quistclose trust arose. 83 In Envy, the

Attorney-General v Aljunied-Hougang-Punggol East Town Council [2015] SGHC 137 [124] ibid [126]. 82 ibid. 83 Envy Asset Management v CH Biovest [2024] SGHC 46, [51] 81

269

Singapore Comparative Law Review

The Quistclose Trap: Intention In Quistclose Trusts

issue was whether investors could recover monies they had paid into a Ponzi scheme that they had falsely believed to be a lucrative nickel trading opportunity. Goh J found that the contracts did not show an intention to subject the funds to the “requisite exclusivity of purpose, restriction against free disposal, or segregation, so as to have given rise to a Quistclose trust”. 84 A clause in the agreement that the investment was to be used “solely for investment in LME Nickel Grade Metal” was “much too vague to have constituted the basis of a specific and exclusive purpose behind the payment”. 85 This is probably correct: investment activities usually require the flexibility of free disposal, such as in entering and exiting the market at the right time. And on proper analysis, the investment vehicle was “never obliged to actually carry out any specific action with the principal amount”. 86 Thus, considering specificity of purpose might sometimes be helpful, but it should not displace the true question of whether the monies were at the transferee’s free disposal. V.

CONCLUSION

It is hoped that this article has been able to dispel much of the controversy around the Quistclose trust. It has been shown that by adopting Lord Millett’s test—that a Quistclose trust arises whenever it is intended by parties that the assets transferred would not be at the free disposal of the transferee—the law can be made more coherent and principled. Moreover, it would benefit the law if it is instead the notion of free disposal that becomes the focus of the Quistclose trust.

ibid [49]. ibid [51]. 86 ibid [50]. 85

270

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

RECOVERING THE BENEFICIARIES’ PROFIT FROM THEIR FIDUCIARY Meng Lingyi* ABSTRACT This essay examines whether an account of profits for breach of fiduciary duty should require proof that the fiduciary's breach was a but-for cause of the profit obtained. It analyses the divergence between the Singapore Court of Appeal's decision in UVJ v UVH, which introduced a but-for causation requirement for disgorgement, and the UK Supreme Court's subsequent decision in Recovery Partners GP Ltd v Rukhadze, which rejected counterfactual reasoning in favour of an attributional nexus between the fiduciary position and the profit. The essay argues that the but-for causation rule is both doctrinally unsound and normatively undesirable. Doctrinally, it incorrectly imports the causation principles governing equitable compensation into the distinct, gain-based remedy of an account of profits, despite the fundamentally different purposes served by the two remedies. Normatively, a but-for requirement weakens the prophylactic function of fiduciary law by permitting fiduciaries to retain gains obtained through conflicts of interest whenever they can establish that the same profits would have been earned lawfully. Contrastingly, an attributional approach better reflects the relationship-based nature of fiduciary obligations, preserves the deterrent function of the no-profit rule, and accommodates concerns of fairness through equitable allowances rather than counterfactual causation. The essay concludes that Singapore should develop an attributional causation standard, supported by burden-shifting principles, instead of adhering to UVJ's but-for framework. I.

INTRODUCTION

When a fiduciary breaches the duty of loyalty by exploiting a conflict of interest or profiting from their position, equity's traditional response is to compel disgorgement through an account of profits. Fiduciary law is built upon an obligation of "single-minded loyalty," and the profit and conflict rules exist to protect and deter those who have undertaken such an obligation from being tempted by human frailty to fall short of it. The stringency of that response has, however, generated a contested causation question: must the fiduciary's breach be a but-for cause of the

King's College London, LLB (Hons), Class of 2027. I am profoundly grateful to the reviewers from the editorial team for their comments on earlier drafts of this article. All errors that remain are my own. E-mail for correspondence: lingyim2627@gmail.com.

271

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

profit before disgorgement can be ordered, or does a looser nexus between the profit and the fiduciary position suffice? The question has assumed renewed doctrinal urgency in light of two significant apex court decisions delivered in 2020 and 2025 respectively, which have reached materially different answers. In Singapore, the Court of Appeal in UVJ v UVH1 held that profits sought to be disgorged via an account of profits "must be caused by" the breach of fiduciary duty, warning against recovery of profits that are insufficiently connected to the wrong committed. That formulation, at least on its face, imports a but-for causation requirement into the gain-based account. In England, by contrast, the UK Supreme Court in Recovery Partners GP Ltd v Rukhadze2 emphatically rejected the use of any but-for counterfactual in identifying accountable profits. Lord Briggs, delivering the leading majority judgment, held that the operative question is not whether the profit would have been made even without any antecedent breach of fiduciary duty, but rather whether the profit owed its existence to a significant extent to the application by the fiduciary of property, information or some other advantage enjoyed by reason of their fiduciary position, or from conflicted activity undertaken whilst still a fiduciary. For that purpose, the court looks closely at the facts as they actually occurred and does not concern itself with what might have happened in a hypothetical but-for situation that did not in fact occur. This essay argues that UVJ's but-for causation requirement is doctrinally unsustainable and normatively inferior to the attributional nexus approach, and that Singapore law should develop a principled account of attributional causation complemented by burden-shifting and equitable allowance. The essay proceeds as follows. Section 2 sets out the foundational doctrinal framework: beginning with the nature of the duty of loyalty, before describing the Singapore position through the lens of Winsta and UVJ and the English position as restated by the Rukhadze majority. Section 3 analyses UVJ's reasoning on three grounds: Lord Briggs’s repudiation of UVJ; UVJ's analogy between equitable compensation and account of profits; and the two justifications offered by Winsta applied in the disgorgement context. Section 4 advances the normative case for attributional causation, drawing on the psychology of selfdeception in fiduciary decision-making, the institutional signalling function of the no-profit rule, and the divergence between Singapore and English law as a reflection of deeper

1 2

[2020] SGCA 49 [2025] 2 WLR 529

272

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

disagreement about what fiduciary law is for. Section 5 addresses how Singapore law ought to proceed, considering the allocation of the burden of proof, the case against premature legislative codification, and the comparative Commonwealth consensus. Section 6 offers a conclusion. II.

THE DOCTRINAL FRAMEWORK OF THE DUTY TO ACCOUNT A. The Nature of the Duty to Account

The starting point for any analysis of causation in the account of profits context may be the nature of the duty whose breach gives rise to the remedy. The duty of loyalty is the defining obligation of a fiduciary relationship, and its character bears directly on what remedial logic is appropriate when it is breached. Millett LJ's celebrated statement in Bristol and West Building Society v Mothew, 3 adopted as the authoritative formulation in Rukhadze, describes the duty of loyalty as requiring a fiduciary to act in good faith, not to make a profit out of their trust, not to place themselves in a position where duty and interest may conflict, and not to act for their own benefit or the benefit of a third person without the informed consent of their principal. The conflict and profit rules, the no-conflict and no-profit duties, are facets of this single overarching obligation of loyalty. Crucially, those rules are prophylactic rather than purely compensatory in their design. Lord Briggs in Rukhadze explained the essentially prophylactic role of the conflict and profit rules as being to protect and deter those who have undertaken an obligation of single-minded loyalty from being tempted by human frailty to fall short of that obligation, tracing this principle back to Keech v Sandford4. In that case, a trustee who renewed a lease for his own benefit, after the landlord refused to renew for the infant beneficiary, was held liable to account. Lord King LC reasoned that if a trustee might have a lease to himself upon the landlord's refusal, "few trust estates would be renewed to cestui que use," and that "it is very obvious what would be the consequence of letting trustees have the lease, on refusal to renew to cestui que use." The rule applied notwithstanding that the trust could not in any event have obtained the benefit. The logic is explicitly deterrence-based: strict liability is imposed not because the principal suffered

3 4

[1998] Ch 1 (1726) Sel Cas Ch 61

273

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

a causally linked loss, but because any other rule would systematically undermine the integrity of the fiduciary relationship. Singapore law adopts the same foundational position. In Sim Poh Ping v Winsta Holding Pte Ltd5, the Court of Appeal affirmed, drawing on Mothew, that one of the core duties of a fiduciary is not to act for their own benefit or that of a third party without the informed consent of the principal, and noted that the no-conflict rule is a prophylactic principle aimed at avoiding the risk that the director might prefer personal or third-party interests over those of the company. B. UVJ UVJ is an important companion case to Winsta, which made a comprehensive restatement of Singapore law on equitable compensation for breach of fiduciary. UVJ addressed the question of causation in the specific context of the account of profits as a gain-based remedy, holding that a breach must be a but-for cause for the accountable profit. In UVJ, three brothers, who were executors of their father’s estate, intentionally kept the Sisters uninformed of their share in companies in the Estate’s assets and failed to distribute them. 6 If causation is unnecessary, such that the profit need only be obtained where there is a possible conflict between the fiduciary’s duty and interest, or by reason of his taking advantage of his fiduciary position, the brother must account for the profit. 7 Conversely, if causation is necessary, as per submitted by the Brothers, they would still obtain the directors’ remuneration which the Sisters could not have interfered with. The court ruled in favour of the brothers, holding that “the profits sought to be disgorged via an account of profits must be caused by the breaches of fiduciary duty, whether this be that the trustee acted in conflict of interest or was guilty of some other breach.” This prevents equity from becoming “an unruly horse where any breach by a fiduciary can be used to recover a profit however unconnected the two may be, and even if the profits would have been earned by the fiduciary in the absence of the breach.”8

[2020] SGCA 35 UVJ (n 1) [66] 7 UVJ (n 1) [77] 8 UVJ (n 1) [98] 6

274

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

Beyond offering the negative reasons of how precedents did not preclude such a test, the court drew an analogy to the measurement of equitable compensation to positively justify the butfor causation test: “no equitable compensation for non-custodial breaches of fiduciary duty can be claimed in respect of loss which the fiduciary can show would have been sustained in spite of the breach”, and “if this investigation can be carried out in situations involving equitable compensation, there is no reason why it cannot be similarly done for an account of profit.”9 It also rejected the argument that deterrence can justify the refusal to investigate hypothetical situations as to what would have happened if the fiduciary had performed his duty. Applying that framework to the facts, the Court found that the estate's minority shareholding made no difference to the resolutions: even had the estate's shares been voted against, the majority shareholders would have passed the resolutions regardless. The claim for an account of profits accordingly failed for want of causation. C. UVJ’s supporting rationales from Winsta Winsta represents the most authoritative restatement of Singapore law on equitable compensation for breach of fiduciary duty to date. It sets out that equitable compensation for non-custodial breach of fiduciary duty cannot be claimed if the fiduciary can show that it would have been sustained despite the breach. 10 This case is of particular importance to the court’s justification in UVJ, as the court imported the investigation of equitable compensation in Winsta by explaining that there is no reason why the same cannot be done for an account for profit. Due to the subject matter of the essay being the causation analysis, it does not discuss the validity of the analysis of Winsta for equitable compensation. It instead focuses on if similar reasoning could be applied to justify applying but-for reasoning to measure the accountable profit for breaches of the duty of loyalty by fiduciaries in UVJ. In Winsta, the Court of Appeal rejected the measures for equitable compensation where causation is never or always relevant. Instead, it adopted a burden-shifting model: once the principal proves breach of a non-custodial fiduciary duty and that loss was suffered, a rebuttable presumption of causation arises, and the fiduciary bears the legal burden of proving that the loss would have occurred in any event. As a result, “causation is relevant but the burden

UVJ (N 1) [88] Winsta (n 5)

275

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

of proof is reversed inasmuch as the defendant will have to prove that the damage suffered by the plaintiff would have occurred in any event.”11 The conclusion is justified on two bases. Firstly, the court propounded that it is illogically to do away with the concept of causation since it is “a logical construct that, ex hypothesi, is of general or universal application and which therefore transcends, inter alia, the divide between equity on the one hand and the common law on the other.” 12 Secondly, the court explained that applying the common law but-for test in equity achieves a more favourable result. The court says that the best approach to resolving issues that involves substantial principle of both common law and equity, “the best approach is one of balance – in particular, the courts do best when they endeavour to embrace, as far as is possible, all the relevant rules and principles of common law and equity, allocating the appropriate ‘legal space to each and utilising them as and when appropriate in order to achieve a just and fair result in the case at hand.”13 D. The English Law position Rukhadze14 arose from the defendants' exploitation of a business opportunity, providing asset recovery services for an oligarch's estate, which they had first encountered in the course of their duties as agents for the claimant companies. After resigning and setting up a competing venture, they earned substantial profits. The claimants sought an account of those profits on the ground that they had been made in breach of the defendants' fiduciary duties of loyalty. The majority of the UK Supreme Court held that the defendants were liable to account for all of the profits earned. The operative test for identifying accountable profits, as stated by Lord Briggs, is whether the profit owed its existence to a significant extent to the application by the fiduciary of property, information or some other advantage enjoyed as a result of the fiduciary position, or to some conflicted activity undertaken whilst the fiduciary remained in post. 15 The court looks at what actually happened; it does not construct a counterfactual scenario in which the breach did not occur.16

Winsta (n 5) [131] Winsta (n 5) [97] 13 Winsta (n 5) [97] 14 Rukhadze (n 2) 15 Rukhadze (n 2) [36] 16 Rukhadze (n 2) [37] 12

276

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

Critically, the majority rejected both species of but-for counterfactual argument that had been advanced. Lord Briggs identified two potential aspects of a "but for the breach" counterfactual. The first, asking what profit the principal would have made but for the breach, had never been in issue, as it was common ground that the account does not operate as a means of equitable compensation for loss.17 The second, which the defendants did seek to introduce, was asking what profits the fiduciary would or might have made for themselves had they committed no breach of duty. The majority rejected this second species of counterfactual as impermissible, endorsing the statement of Lord Radcliffe in Gray v New Augarita Porcupine Mines Ltd [1952] 3 DLR 1 that such an argument should not be heard. The determination of a court of equity to avoid a but-for counterfactual in the identification of accountable profits is, the majority held, firmly established by clear and consistent authority. 18 III.

ANALYSING THE REASONING IN UVJ A. Is Lord Brigg’s Criticism of UVJ Sustained?

A good starting point for the analysis of UVJ's approach to causation in the account of profits is Lord Briggs’s express repudiation of that decision in Rukhadze. In his judgement, Lord Briggs sets out two points of errors of approach in the UVJ dictum of the Singapore Court of Appeal. The first criticism is grounded in Lord Briggs's characterisation of the duty to account as an independent primary obligation rather than a remedial response to a distinct antecedent breach. On Lord Briggs's analysis, “a prior breach is neither necessary nor sufficient to trigger an account of profits, although such a breach may well provide evidence of a sufficient link, and usually a causative link, between the fiduciary's office as such and the receipt of a profit.” 19 This is because the duty to account for profit is an independent duty, not just a remedy for some other breach. On this view, UVJ's entire analytical framework is misconceived: because accountability does not depend on a prior breach, asking whether the breach caused the profit is the wrong question from the outset. Secondly, a but-for type of causation requiring the erection of a counterfactual is not necessary even if a causative link is material to the identification of accountable profits. On the facts of

Rukhadze (n 2) [37] Rukhadze (n 2) [38] 19 Rukhadze (n 2) [66] 18

277

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

UVJ, since the brothers were re-appointed as remunerated directors of the three companies by the votes of the majority shareholders, the estate’s tiny minority of shares played no significant part in that outcome. Arriving at this conclusion does not require the erection of counterfactual.20 Lord Briggs's second criticism follows logically from the first. Permitting a fiduciary to resist disgorgement by proving they could have made the profit anyway would water down the primary duty to account and reintroduce the speculative inquiry that equity has historically refused, on the ground that the safety of mankind requires that no agent be put their principal to the danger of such an inquiry. 21 In this analysis, the but-for counterfactual actively undermines the prophylactic function of the rule. However, Lord Briggs's critique does not command universal assent within Rukhadze itself. Lord Leggatt expressly rejects the "duty theory", the characterisation of the duty to account as an independent primary obligation, as both unnecessary and conceptually unsound, and as having no support in the authorities. 22 On Lord Leggatt's analysis, the liability to account for unauthorised profits is properly understood as a remedial response to a wrong, breach of the duty of loyalty, and ordinary causation principles accordingly apply: the profit must be caused by the breach, and identifying whether it is so caused requires the construction of an appropriate counterfactual. This is UVJ's starting premise: that the account of profits is a remedy for breach and that a prior breach is therefore the trigger for the account. The more defensible position, it is submitted, is that the internal division within Rukhadze reveals two internally coherent but incompatible frameworks. In Lord Leggatt's framework, and UVJ's, the account of profits is a remedy triggered by breach, and some form of causal inquiry linking breach to profit is appropriate. The real debate on this view is not whether causation is relevant but how the causation inquiry should be structured. On Lord Briggs's framework, the prior question, whether a breach occurred, is analytically separable from the primary obligation to account, and the but-for counterfactual is excluded not because causation is irrelevant but because the wrong question is being asked. The implications of choosing between these frameworks for Singapore law are considered further below. B. Flawed Analogy in UVJ: Equitable Compensation and Account of Profits Are Not Equivalent

Rukhadze (n 2) [67] Rukhadze (n 2) 22 Rukhadze (n 2) 21

278

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

Leaving aside the primary duty debate, UVJ's reasoning is vulnerable to a distinct and serious objection: it draws an analogy between equitable compensation and account of profits that is doctrinally unsustainable. The Court of Appeal in UVJ itself emphasised the conceptual distinction between surcharging, a loss-focused, compensation-like response, and an account of profits, a gain-focused response, and cautioned against conflating the two remedies. 23 It reiterated, citing Mona Computer Systems (S) Pte Ltd v Singaravelu Murugan [2014] 1 SLR 847, that an account of profits is a gains-based remedy and is not related to whether the principal suffered loss or would have made the profit but for the breach. 24 Yet, having distinguished the two remedies in principle, the Court then justified introducing a but-for causation requirement into the account of profits by analogy with equitable compensation, reasoning that if counterfactual reconstruction is achievable for equitable compensation, there is no reason why it cannot similarly be undertaken for an account of profits. The analytical inconsistency is evident as UVJ warns against conflating compensation and disgorgement, then uses the methodology of the former to determine the scope of the latter. Lord Briggs makes the same point in Rukhadze, describing equitable compensation and account of profits as being "like chalk and cheese": equitable compensation is about compensation for loss, whereas loss is irrelevant to an account of profits.25 The two remedies serve fundamentally different juridical purposes. Equitable compensation seeks to repair the loss suffered by the principal, making causal analysis indispensable because liability is measured by the claimant's detriment. The but-for test is appropriate here because it measures the difference between the world in which the breach occurred and the world in which it did not to calculate what must be made good. An account of profits, by contrast, is concerned with the fiduciary's unauthorised gain rather than the principal's loss. It is irrelevant whether the plaintiff suffered any loss at all, and it is likewise irrelevant that the fiduciary acted in good faith. The rule is best understood as a relationship-based rule of attribution rather than a response to harm.26 The rationale of the account of profits is therefore prophylactic rather than compensatory. Lord Briggs in Rukhadze anchors the profit and conflict rules in their prophylactic and deterrent purpose: protecting fiduciary single-minded loyalty by removing the temptation arising from human frailty. The account of profits gives effect to that purpose by ensuring that no financial

UVJ (n 1) UVJ (n 1) 25 UVJ (n 1) 26 UVJ (n 1) 24

279

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

benefit can be retained from a conflicted position. 27 Requiring the court to reconstruct a hypothetical but-for world shifts the inquiry away from whether the fiduciary abused a position of trust towards whether the fiduciary would ultimately have obtained the same profit through some other means. In doing so, the remedy begins to resemble damages rather than disgorgement. Because the beneficiary's entitlement is framed as a primary right of attribution of the gain actually made, counterfactual arguments, that the fiduciary could have made the same gain lawfully or that informed consent would have been given, are treated as irrelevant; importing such counterfactualism wrongly makes the claim look like damages for wrongdoing.28 C. Winsta's Two Justifications for But-For Causation and Their Limits in the Disgorgement Context UVJ's reasoning draws implicitly on the framework adopted for equitable compensation in Sim Poh Ping v Winsta Holding Pte Ltd [2020] SGCA 35. However, the burden-shifting framework was justified on two bases, each of which, when transposed to the account of profits context, is open to serious objection. Is a Counter-Factual Logically Necessary? The first justification was that causation is a logical construct of general or universal application transcending beyond the divide between equity and common law, making a nocausation approach difficult to sustain as a matter of principle. 29 UVJ drew on this logic to argue that there is no principled reason to exempt the account of profits from causation analysis. The difficulty with this argument when applied to the account of profits is that it conflates two fundamentally different things: the general proposition that some connection between breach and outcome is required, and the specific claim that this connection must take the form of a counterfactual but-for inquiry. Lord Briggs in Rukhadze draws this distinction, separating common law causation from a broader factual "A led to B" inquiry implicit in phrases such as "by reason of," "out of," and "owing to," which does not require positing counter-factuals.30 On his account, the relevant inquiry for the account of profits can be purely actualistic; “did

UVJ (n 1) UVJ (n 1) 29 Rukhadze (n 2) 30 Rukhadze (n 2) 28

280

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

the profit owe its existence significantly to the fiduciary position and advantages?”, without constructing hypothetical scenarios in which the breach did not occur. 31 Causation as a logical construct does not, therefore, entail the but-for test as its exclusive or even primary expression. The universality argument proves only that some nexus is required; it does not determine what form that nexus should take, and in the profit-stripping context, there are strong reasons, examined above, for preferring a non-counterfactual attribution inquiry over a but-for reconstruction. The Right Balance between Equity and Common Law The second justification offered in Winsta was that the best approach to issues involving both common law and equity is one of balance, with courts endeavouring to embrace all relevant rules and principles and allocate appropriate legal space to each in order to achieve a just and fair result.32 UVJ deployed this reasoning to argue that importing but-for causation from the common law damages framework into the account of profits achieves a fairer outcome by preventing recovery of profits that are entirely unconnected to any breach. This justification is, however, inadequate in the disgorgement context for two related reasons. First, the "balance" that Winsta advocates is between equitable compensation and common law damages, both of which are loss-based, reparative remedies. The question of "balance" in that context is about calibrating how strictly a causal requirement should be applied before loss is compensated. The account of profits is not a reparative remedy but prophylactic and gainbased. Reasoning that assumes the same causation inquiry should govern equitable compensation and account of profits risks collapsing distinct juridical functions into a single damages-style analysis.33 Second, and more fundamentally, the prophylactic purpose of the account of profits means that fairness in this context cannot be assessed solely by matching outcomes to counterfactual financial positions. The rule's fairness is institutional as much as transactional as it preserves the integrity of the fiduciary relationship and deters disloyalty through clear, strict, and predictable rules.34 As Samet argues, allowing fiduciaries to argue that their breach caused no

Rukhadze (n 2) UVJ (n 1) 33 Lionel Smith, The Law of Loyalty (OUP 2023) 196. 34 Rukhadze (n 2) 32

281

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

real harm, or that they would have made the same profit in any event, gives rise to vulnerability to self-deception and biased judgment, which is what the categorical rule was designed to foreclose. A balance that dilutes that strictness in the name of common law coherence accordingly prioritises doctrinal tidiness at the cost of the rule's normative function. 35 D. Singapore Law's Current Position: Open Questions Drawing together the threads above, Singapore law's current position on the causation requirement for account of profits is best understood as a developing framework that has established a principled starting point in UVJ, whilst leaving a number of important questions deliberately open for resolution in future cases. UVJ held that profits sought via an account of profits must be caused by the fiduciary's breaches, warning against recovery of profits that are however unconnected to the breach, even if the profits would have been earned in the absence of the breach. On the facts, the Court of Appeal found that the Estate's minority shares made no difference to the remuneration resolutions, rejected the non-disclosure theory as an insufficient basis for causation, and accordingly held that the claim for directors' remuneration failed. The holding is clear and is consistently directed at a legitimate concern: that the account of profits should not become an instrument for recovering gains that have no genuine connection to the wrong committed. Two significant questions, however, remain expressly or implicitly reserved for future resolution: The first is the question of the burden of proof. UVJ expressly declined to decide which party bears the burden of proving causation in an account of profits claim, leaving the point open for a future case. One possibility is that Singapore courts will, by analogy with the burden-shifting model adopted for equitable compensation in Winsta, hold that once breach and profit are established, the evidential burden shifts to the fiduciary to disprove the causal connection. This is a reversal broadly consistent with the prophylactic purpose of fiduciary law and with the fiduciary's superior access to the relevant evidence. Another possibility is that the principal must affirmatively establish but-for causation on the balance of probabilities. The choice between these approaches could produce materially different outcomes on the same facts, and

Irit Samet, ‘Guarding the Fiduciary’s Conscience’ (2008) 29 OJLS 775

282

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

calibrating the burden of proof to the remedial purpose of the account of profits is therefore a question of some doctrinal significance. The second question, is one of characterisation: whether the but-for language in UVJ operates as a true counterfactual inquiry, of the kind Lord Leggatt articulated in Rukhadze, asking what the fiduciary would have done had there been no breach, or as a scoping device ensuring that there is a sufficient causal nexus between the fiduciary's position and the profit. The attribution model adopted in Mona Computer Systems (S) Pte Ltd v Singaravelu Murugan [2014] 1 SLR 847 suggests that the limiting principle is attribution within the scope and ambit of the duty of loyalty, framed as a nexus requirement rather than a strict counterfactual test. If so, there is arguably less distance between Singapore law and the Rukhadze majority's sufficient nexus test than UVJ's language of "but-for causation" might initially imply. IV.

THE NORMATIVE CASE FOR ATTRIBUTIONAL CAUSATION

There remains the normative question of which of a but-for causation requirement or an attributional nexus test better satisfies policy considerations. A. Self-Deception and the Deterrent Function of Strict Profit-Stripping The normative case for attributional causation gains further support from a consideration of the psychological dynamics of fiduciary decision-making and the particular risk of self-deception that the no-profit rule is designed to address. The strict profit-stripping rule deters self-deceptive behaviours that undermines proper performance of the trustee's fiduciary duty and the relationship of trust. The administrator's power tends to be discretionary: within the objective requirements of a trust or fiduciary relationship, the fiduciary chooses how to exercise their power subjectively. 36 Even where the action can be objectively characterised as reasonable and honest, the ordinary process of evidence gathering and decision-making may be distorted by self-interest in ways that are invisible to the fiduciary themselves. Self-deception may manifest when virtuous people justify dishonest actions by discounting risks against principles, or when experts subconsciously overemphasise considerations favourable to a desired conclusion so as to arrive at a judgment that serves their interests. These impairments are insidious because they operate below the level of

Lionel Smith, The Law of Loyalty (OUP 2023) ch 3 ('Loyalty in the Exercise of Private Law Powers of Administration').

283

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

conscious awareness; the fiduciary who tells herself that a particular profit-making opportunity is "harmless" or that the principal would certainly have consented may be engaging not in honest assessment but in motivated reasoning. 37 The principal's dependency and informational disadvantage, the combination of discretionary power, informational asymmetry, and the principal's vulnerability, means that principals are particularly susceptible to exploitation of this kind, all the more so because the self-deceived fiduciary presents the appearance of acting in good faith. The strict no-profit rule addresses this problem by removing the opportunity within which self-deceptive reasoning can operate. By making it categorically impermissible to profit from the fiduciary position without informed consent, the rule eliminates the pre-emptive calculation that might otherwise tempt the fiduciary. A but-for causation requirement, even in the burden-shifted form, reopens that same deliberative space. If the fiduciary can resist or reduce disgorgement by establishing that the profit would have been made in any event, or that the principal would have consented to a profit-sharing arrangement, then the pre-emptive calculation becomes rational, as a fiduciary contemplating a conflict may factor into their deliberations the likelihood of being able to run a successful causation defence. As Lady Arden observed in Murad v Al-Saraj38, the temptation to act disloyally is "assisted" if the fiduciary can contemplate escaping liability by arguing no loss was caused or that the profit was not available to the principal. Samet's analysis shows that psychological reasons amplify this effect. The availability of the loophole enables the selfdeceptive rationalisation that leads to disloyalty in the first place. The injustice is further compounded by the fact that the beneficiary bears the full risks should the fiduciary fail to make an honest judgement, while the fiduciary, absent a strict rule, enjoys the benefit and is not under an obligation to seek consent. 39 The attributional causation test avoids this moral hazard. Under an attributional framework, the relevant question is whether the profit was obtained by reason of the fiduciary position, a question directed entirely at what actually happened, with no counterfactual room for the fiduciary to occupy. The deterrent signal is clear and categorical, such that any profit obtained

Samet, ‘Guarding the Fiduciary’s Conscience’ (n 35) 775 [2005] EWCA Civ 959 39 Samet, ‘Guarding the Fiduciary’s Conscience’ (n 35) 775 38

284

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

through the fiduciary relationship must be disgorged, subject only to equitable allowance for legitimate contribution. B. The Institutional and Signalling Function of Fiduciary Law The attributional approach is also consistent with the broader institutional function of fiduciary law. Fiduciary obligations arise because principals entrust discretionary power to another person under conditions of informational asymmetry, and the combination of that power and that asymmetry which both creates the relationship and necessitates its regulation. Courts therefore intervene not only to correct individual instances of unfairness, but also to preserve confidence in relationships of trust more generally. As Smith argues on his attribution account, the no-profit rule is best understood as a rule allocating the benefit of gains obtained through the administrative role to the beneficiary, a rule that reflects the structure of the relationship rather than solely responding to wrongdoing or harm. The rule's institutional function is to signal, clearly and predictably, that those who occupy positions of fiduciary power cannot privately retain benefits generated through those positions. A rule requiring strict accountability for gains obtained through fiduciary office reinforces public confidence that fiduciary powers will be exercised with undivided loyalty. By contrast, a but-for causation requirement may produce outcomes that appear fair to the individual fiduciary, particularly in cases where the fiduciary contributed substantially to the profit, whilst gradually weakening the institutional assurance that underlies the fiduciary relationship more broadly. As Lord Briggs observed in Rukhadze, "the inevitability of accounting (subject only to an uncertain equitable allowance or election for compensation)" is the principal disincentive to entering conflict situations without informed consent. Relaxing the rule through the introduction of counterfactual defences would "water down" the simple duty to avoid such situations. Once fiduciaries can expect that full disgorgement will not follow from all conflicted profit-making, because the outcome depends on a contested but-for inquiry, the categorical assurance that the rule provides is compromised. On the other hand, attributional causation is not equivalent to unlimited liability. It falls short of being “an unruly horse” which enables a principal to receive profits however unconnected to the breach, as described in UVJ.40 Equity has long recognised limiting principles that prevent the account of profits from becoming punitive. A profit must still be sufficiently connected to 40

UVJ (n 1) [98]

285

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

the fiduciary position; purely independent gains remain outside the scope of disgorgement. Likewise, courts retain discretion to award equitable allowances where the fiduciary's own skill, labour, or expenditure has contributed to generating the profit. As Conaglen explains, "just allowances", including for skill and effort, form part of calculating the net profit to be disgorged; if granted, they reflect reasonable and fair remuneration for the work done rather than a concession that the profit was not attributable to the fiduciary position. 41 These doctrines demonstrate that the no-profit rule is not absolute in its practical operation, notwithstanding its strict formulation. They moderate the consequences of disgorgement without requiring courts to abandon the central principle that profits generated through fiduciary offices belong, in equity, to the principal rather than to the fiduciary. C. What the Divergence Reveals About Each Legal System The importance Singapore law places on the individual fairness to the fiduciary, which underlies the case for a but-for requirement, reflects a divergence in how Singapore and English law rank the values at stake in fiduciary adjudication. The fairness argument for a but-for requirement is at its most compelling in cases where the fiduciary's conduct, though technically in breach of the duty of loyalty, generated profits that bear only a marginal connection to the disloyalty itself. On the facts of Rukhadze, the defendants invested considerable work and skill in an asset recovery operation, and the trial judge ordered them to account for net profits reduced by a 25% equitable allowance reflecting that contribution.42 The connection between their antecedent breach of fiduciary duty and the profits ultimately earned was, on one view, relatively attenuated. To require them to disgorge substantially the entirety of those profits, subject only to an equitable allowance whose quantum was uncertain and contested, produces an outcome that superficially resembles punishment rather than equitable correction. The but-for causation requirement has appeal here as it confines disgorgement to the profits that were limited to the product of the wrong, leaving the fiduciary in possession of what they could have earned through entirely lawful conduct. That is arguably a fairer outcome between the parties. UVJ reflects this solicitude for the fiduciary's position at the level of causation doctrine. The Singapore Court of Appeal's insistence that the breach must be a but-for cause of the profit is

41 42

Matthew Conaglen, 'Identifying the Profits for Which a Fiduciary Must Account' (2020) 79(1) CLJ 38 Rukhadze (n 2) [13]

286

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

best understood as a manifestation of a broader judicial commitment in Singapore to ensuring that equitable remedies are calibrated to the wrong committed, that equity does not operate, as the Court put it, as an "unruly horse" enabling a principal to receive profits however unconnected to the breach. 43 This concern is recognisably part of Singapore's general judicial philosophy in the development of equity — a consistent tendency to subject equitable doctrines to principled analytical scrutiny through an integrated and holistic approach that allocates appropriate legal space to both common law and equity, rather than treating equitable doctrines as automatically insulated from rationalist examination. 44 The application of a but-for requirement to account of profits is therefore not an ad hoc departure from fiduciary tradition, it is a deliberate expression of the view that equitable remedies, like their common law counterparts, should be proportionate to the harm or wrong with which they are engaged. English law after Rukhadze, by contrast, reflects a different weighting. Lord Briggs's majority judgment is explicit that the "virtual certainty” of accounting, as a deterrent against entering conflict situations without consent, is the primary value that the no-profit rule serves,45 and that introducing counterfactual defences would "water down" the simple duty to avoid such situations.46 The English approach accordingly prioritises the integrity of the fiduciary relationship and the categorical clarity of the deterrent rule over the individual fairness of outcomes in specific cases. This reflects a conception of fiduciary law in which the relationship, and the public confidence that underpins it, takes precedence over the transactional interests of the individual parties. Fairness to the fiduciary is accommodated not at the liability stage, through causation limits, but at the remedy stage, through equitable allowance and the discretion to award compensation in lieu of an account. The individual fiduciary who contributed legitimately to generating the profit is not left wholly unprotected, but the protection is provided without disturbing the categorical nature of the underlying obligation. Which of these two approaches is to be preferred? It is submitted that, in the fiduciary context, deterrence and institutional integrity should take precedence over individual transactional fairness, and for reasons that are inherent in the structure of the relationship rather than merely consequentialist.

Rukhadze (n 2) [89], [98] Winsta (n 5) [98] 45 Rukhadze (n 2) [59] 46 Rukhadze (n 2) [53] 44

287

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

The case for prioritising fairness to the fiduciary rests on an implicit analogy with other areas of private law in which remedies are calibrated to the wrong. But that analogy is not well founded in the fiduciary context, because the wrong, a breach of the duty of loyalty, is not primarily defined by the harm it causes. Equity intervenes not so much to recoup a loss suffered by the plaintiff as to hold the fiduciary to, and vindicate, the high duty owed. 47 The no-conflict and no-profit rules do not prohibit conduct because it tends to cause loss; they prohibit it because it creates conditions of divided loyalty that are incompatible with the relationship's foundational structure. The fiduciary who enters a conflict of interest has already compromised the relationship regardless of whether any identifiable harm results. As Samet argues, beyond deterring harmful conduct, the strict profit-stripping rule removes the opportunity in which the fiduciary might otherwise engage in the self-deceptive rationalisation that makes disloyalty more likely.48 Calibrating the remedy to the profit that was really caused by the breach enlivens contemplation for breaches before the it occurs. The residual fairness concern, that the fiduciary who made a genuine contribution to generating the profit should not be required to disgorge everything, is, moreover, adequately addressed by equitable allowance without importing counterfactual reasoning into the liability stage. Equitable allowance is not a concession that the profit was not attributable to the fiduciary position, but a recognition that, in calculating the net sum to be disgorged, a just allowance of the fiduciary's legitimate expenditure and contribution should be set against the gross gain.49 Hence, while equitable allowance operates within the attribution framework, a but-for causation reduction operates outside it by denying that the fiduciary relationship generated the profit at all. The former respects the structural logic of the remedy, the latter compromises it. What this analysis ultimately reveals is that the divergence between Singapore and English law is, at its deepest level, a divergence about what fiduciary law is for. Singapore's approach, as expressed in UVJ, treats fiduciary liability as subject to causal limits that ensure proportionality between the breach and the remedy, confining disgorgement to profits causally connected to the wrong and refusing to extend liability to profits the fiduciary would have earned irrespective of any breach. 50 English law's approach, as expressed in the Rukhadze majority, treats fiduciary liability as the enforcement of a categorical, relationship-constituting obligation

Winsta (n 5) [247] Samet, ‘Guarding the Fiduciary’s Conscience’ (n 35) 775 49 Conaglen (n 41) 42 50 UVJ (n 1) [88] 48

288

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

whose function is irreducibly institutional and cannot be reduced to individual corrective justice without distorting its character. 51 On the first understanding, individual fairness and causal proportionality are organising values; on the second, the integrity of the fiduciary relationship and the deterrent efficacy of the categorical rule take precedence. These two understandings reflect different conceptions of what makes fiduciary obligations distinctive and why they merit the special remedial treatment that equity affords them. It is submitted that second understanding is more faithful to the historical and normative foundations of fiduciary law. The account of profits is not, at root, a remedy calibrated to individual harm. It is the legal expression of a relational commitment, enforced categorically to preserve the integrity of relationships upon which principals must be able to rely without reservation. Singapore law itself recognises that the no-conflict and no-profit rules are strict in character and do not depend on fraud or bad faith, and that fiduciaries can avoid apparent harshness by obtaining the principal's informed consent before acting. 52 A causation standard that returns the fiduciary to the position they would have occupied through lawful conduct fails to vindicate that commitment, because it treats the fiduciary's hypothetical loyal conduct as a baseline from which the wrong is measured, rather than treating the relationship of loyalty itself as the baseline from which any departure must be accounted for in full. V.

ADDITIONAL CONSIDERATION ABOUT WHAT THE DOCTRINAL TEST SHOULD BE

The conclusions of the preceding sections raise a final set of questions about how Singapore law ought to proceed. The case for legislative codification of a specific causation standard for fiduciary breach might initially seem attractive as it would provide certainty and resolve the tensions left open by UVJ. However, codification would at this stage be premature, with the law of causation in equitable compensation being historically been clouded by confusion and requiring careful calibration. 53 The appropriate response to the current doctrinal uncertainty may not be to crystallise one approach legislatively before its implications have been fully worked through judicially, but to allow the courts to continue developing the law in a remedysensitive manner. This is particularly so because, as this essay has argued, the appropriate causation standard for equitable compensation and that for account of profits are not the same

Rukhadze (n 2) [47], [58] Winsta (n 5) [251] 53 Winsta (n 5) [130] 52

289

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

and should not be governed by a single unified rule. A uniform legislative test risks entrenching the very confusion it seeks to resolve. The most immediately pressing question that UVJ left open is the allocation of the burden of proof. The Court of Appeal held that profits must be caused by the fiduciary's breach but expressly reserved the question of which party bears the burden of establishing that causal connection.54 The allocation of the burden may prove outcome-determinative on many sets of facts. The preferable resolution is that Singapore courts should transpose Winsta's burdenshifting framework into the account of profits context, with appropriate modifications to reflect the gains-based character of the remedy. Under this transposed framework, the principal would bear the legal burden of establishing the attributional nexus, that the profit was obtained by reason of the fiduciary position. Once that nexus is established, the burden should shift to the fiduciary to prove that the profit falls outside the scope of the duty to account or that an equitable allowance ought to be granted. This allocation is justified on both normative and practical grounds. Normatively, it reflects the principal's dependency and vulnerability: fiduciaries are often better placed to show how loss was or was not occasioned, and the rule supports deterrence and addresses information asymmetry. Practically, a fiduciary given wide discretion to make decisions and enter into transactions will likely be in a better position to show how the profit was generated, as well as to meet the threshold of proving that it would have arisen even without the breach. Placing the burden on the fiduciary at the second stage accordingly corrects the evidential imbalance that would otherwise disadvantage the principal without undermining the logical structure of the attribution inquiry. On the broader comparative question, this essay has argued that the Rukhadze majority's sufficient nexus test is normatively superior to UVJ's but-for requirement in that it better preserves the prophylactic character of fiduciary law and avoids the moral hazard introduced by counterfactual deliberation. It does not follow, however, that Singapore should simply adopt the Rukhadze framework wholesale. The internal division within Rukhadze, between Lord Briggs's primary duty theory, Lord Leggatt's minimum departure counterfactual, and Lord Burrows's intermediate position, means that English law is itself unsettled. The preferable course for Singapore is to develop its own principled account of attributional causation,

UVJ (n 1) [98]

290

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

drawing on the Rukhadze majority's insights without being bound by a framework whose foundational premises remain contested even within the court that produced it. Such an approach would also align Singapore more closely with a broader Commonwealth consensus such as Canadian and Australian, where courts have generally treated causation as relevant in the fiduciary context but have adopted burden-shifting approaches that place the onus on the fiduciary to rebut the causal connection. 55 That consensus points towards a model that is simultaneously more faithful to the prophylactic purpose of fiduciary law than UVJ's strict butfor requirement and more analytically coherent than the unresolved plurality in Rukhadze. VI.

CONCLUSION

This essay has argued that UVJ's but-for causation requirement is both doctrinally unsustainable and normatively counterproductive. The doctrinal objection operates on two levels. First, the Rukhadze majority correctly identifies that the account of profits operates by reference to what actually happened, whether the profit owed its existence to a significant extent to the fiduciary's position and advantages, and that this inquiry is purely actualistic, requiring no construction of a hypothetical world in which the fiduciary acted loyally. Second, and more fundamentally, UVJ's own reasoning is internally inconsistent: having correctly distinguished equitable compensation from account of profits as conceptually distinct remedies, the former reparative and loss-focused, the latter prophylactic and gain-focused, the Court of Appeal then justified importing a but-for causation requirement into the account of profits by analogy with the compensatory framework it had distinguished. As Lord Briggs observed, the two remedies are "like chalk and cheese," and the methodology of the former cannot be transposed into the latter without distorting the remedial logic of both. Winsta's two justifications for a burden-shifted but-for inquiry fare no better when applied to the disgorgement context: causation as a logical universal demands only some nexus between breach and profit, not that the nexus must take counterfactual form; and the balance argument transposes the reasoning of a reparative context into a prophylactic one, importing the logic of the loss-based damages framework into a remedy whose function is irreducibly different. The normative case for attributional causation rests on grounds that are structural instead of consequentialist. The strict profit-stripping rule addresses the intractable problem that a fiduciary who enters a conflict of interest may engage in self-deceptive reasoning, convincing 55

Brickenden v London Loan & Savings Co of Canada [1934] 3 DLR 465, Warman International v Dwyer (1995) 182 CLR 544

291

Singapore Comparative Law Review

Recovering The Beneficiaries’ Profit From Their Fiduciary

herself that the profit is harmless or that consent would have been given, in ways that are invisible to the decision-maker herself and not amenable to ordinary proof. A but-for causation requirement, even in the burden-shifted form that a Winsta transposition might produce, reopens the deliberative space that the categorical rule was designed to foreclose: the fiduciary who can contemplate a successful causation defence has reason to perform the motivated reasoning that the no-profit rule is designed to preclude. The divergence between UVJ and Rukhadze ultimately reflects a deeper disagreement about the normative character of fiduciary law. Singapore's approach treats fiduciary liability as a species of private law liability generally, subject to causal limits that ensure proportionality between breach and remedy, a conception in which individual corrective justice and causal proportionality are organising values. English law, as expressed in the Rukhadze majority, treats fiduciary liability as the enforcement of a categorical, relationship-constituting obligation whose function cannot be reduced to individual corrective justice without distorting its character. This essay has argued that the latter understanding is more faithful to the historical and normative foundations of fiduciary law. For Singapore law, the appropriate path forward is not the wholesale adoption of the Rukhadze framework, whose own internal premises remain contested, but the principled judicial development of an attributional causation standard. That standard would require the principal to establish the attributional nexus, that the profit was obtained by reason of the fiduciary position, at which point the burden would shift to the fiduciary to disprove the connection or to establish the grounds for an equitable allowance. This approach is consistent with the prophylactic purpose of the no-profit rule, corrects the evidential imbalance inherent in the fiduciary relationship, and aligns Singapore more closely with the broader Commonwealth consensus. Legislative codification at this stage would be premature; the courts are better placed to develop these nuances in a remedy-sensitive manner. The account of profits is not concerned with a hypothetical world in which the fiduciary acted loyally. It is concerned with what the fiduciary actually did with a position of trust and whether, by reason of that position, a profit was obtained. A causation standard that returns the fiduciary to the position they would have occupied through lawful conduct fails to vindicate the foundational commitment of the fiduciary relationship, because it treats the fiduciary's hypothetical loyal conduct as the baseline rather than treating the relationship of loyalty itself, the undivided, single-minded loyalty owed by every fiduciary, as the baseline from which any departure must be fully accounted for.

292

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

DEFENDING THE QUASI-EMPLOYEE DOCTRINE IN SINGAPORE IN THE AFTERMATH OF BIRD V DP Lam Yu Han* and Sim Suan Yeong, Amelita** ABSTRACT The rule that an organisation may be held vicariously liable for a tort committed by a tortfeasor who is not its traditional employee, but whose relationship is nonetheless “akin to employment” (commonly known as the “quasi-employee doctrine”), has been accepted only in obiter dicta in Singapore. In Bird v DP, the High Court of Australia rejected the quasi-employee doctrine. Bird may therefore provide impetus for the Singapore courts to do the same. This article argues, however, that the quasi-employee doctrine should be retained for four reasons. First, it enables unincorporated associations to be appropriately held vicariously liable. Second, the Singapore courts are institutionally competent to accept the doctrine. Third, the formalistic reasoning adopted by the High Court of Australia in rejecting the quasi-employee doctrine in Bird is inconsistent with Singapore’s existing multifactorial approach to determining employment relationships. Fourth, the quasi-employee doctrine ensures that vicarious liability remains relevant in the contemporary era. I. INTRODUCTION 1. In Bird v DP,1 the claimant was sexually assaulted by an assistant priest (deceased at the time of hearing) belonging to one of the churches under the purview of the Roman Catholic Diocese of Ballarat (“the Diocese”).2 The claimant sought compensation from the Diocese for the psychological harm done to him as a result of the sexual assault by arguing that the Diocese was vicariously liable for the assistant priest’s act.3 The majority of a seven-judge panel of the High Court of Australia (“HCA”) denied the claim, holding that only a “strict” employment relationship would suffice for the imposition of vicarious liability. 4 It held that

Singapore Management University (SMU) (LLB Class of 2028). We are grateful to the reviewers from the editorial team for their comments on earlier drafts of this article. All errors that remain are our own. ** National University of Singapore (NUS) (LLB Class of 2029). 1 Bird v DP (a pseudonym) [2024] HCA 41 (“Bird”). 2 ibid [1]. 3 ibid [2]. 4 ibid [47].

293

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

the priest was not under a relationship of employment with the Diocese 5 and therefore the Diocese could not be held vicariously liable for the acts of the priest. 6 2. Bird represents a rejection of the quasi-employee doctrine in vicarious liability. 7 This doctrine sets out that vicarious liability can be found for a tort committed by a tortfeasor which does not have a traditional employment relationship with the organisation, but has a relationship that is nonetheless “akin to employment”. The quasi-employee doctrine is now settled law in the UK, established in Various Claimants v Catholic Child Welfare Society. 8 The facts there were very similar to Bird. An organisation known as the Brothers of the

Christian Schools (or “the Institute”) ran schools for children. 9 The claimants alleged they were sexually abused by the teachers in a school (“St Williams”) 10 which was under the de facto control of the Institute. 11 An issue of whether the teachers were employees of the Institute arose because the teachers did not contract with the Institute, but with the management of St Williams.12 The UK Supreme Court (“UKSC”) applied the quasiemployee doctrine13 and held that the Institute was vicariously liable. 14 The BrothersInstitute relationship was held to be sufficiently akin to employment. The reasons given were: the Institute was structured akin to a corporate body with a clear hierarchical structure;15 the Brothers taught pursuant to the Institute’s direction; 16 their teaching was carried out in furtherance of the Institute’s mission;17 and the manner in which they performed their work was governed by the Institute’s rules. 3. Bird potentially jeopardises the applicability of the quasi-employee doctrine in Singapore. While the quasi-employee doctrine was acknowledged with approval by the Singapore Court of Appeal (“CA”) in Ng Huat Seng v Munib Muhammad Madni,18 this was merely obiter dicta. The claim there was denied on the basis that the tortfeasor was an independent

ibid [11]. ibid [72]. 7 ibid [47]. 8 Various Claimants v Catholic Child Welfare Society and others [2013] 2 AC 1 (UKSC) (“Christian Brothers”). 9 ibid [1]. 10 ibid [4]. 11 ibid [25] and [44]. 12 ibid [56]. 13 ibid [94]. 14 ibid [4]. 15 ibid [89]. 16 ibid [89]. 17 ibid [89]. 18 Ng Huat Seng v Munib Mohammad Madni [2017] 2 SLR 1074 (“Ng Huat Seng”) at [63]. 6

294

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

contractor engaged by the defendant. 19 The quasi-employee doctrine was thus, strictly speaking, not engaged in Ng Huat Seng. 4. It is acknowledged that since Ng Huat Seng, the High Court (“HC”) in Ong Han Ling v American International Assurance Co Ltd 20 and Munshi Mohammad Faiz v Interpro Construction21 has applied the quasi-employee doctrine to hold that a qualifying relationship existed to find vicarious liability notwithstanding the absence of traditional employment. In Ong Han Ling, the court held that, although the tortfeasor-agent “may not be [technically] considered AIA’s employee” on the face of the contractual arrangement between the parties,22 the relationship was sufficiently analogous to employment, in that it possessed the same fundamental qualities as an employer-employee relationship, to justify the imposition of vicarious liability. 23 In Munshi, vicarious liability was imposed on the same basis 24 notwithstanding the absence of any contract of employment between the tortfeasor and the first defendant. 25 5. However, Ong Han Ling and Munshi may not be the last word on the matter. Both cases relied on Ng Huat Seng’s obiter acknowledgement of the quasi-employee doctrine as authority for its applicability in Singapore. 26 Further, in the wake of Bird (which was decided after Ong Han Ling and Munshi), it is possible that a defendant in a vicarious liability case will attempt to invoke Bird as persuasive authority in a case where a relationship of employment is not explicit on the facts. Hence, it remains possible that if a case involving the quasi-employee doctrine were to come before the HC or CA, the court might rely on Bird rather than Christian Brothers and conclude that only a traditional employment relationship gives rise to vicarious liability. 6. Relying on Bird however may result in a denial of redress for an otherwise deserving claimant due to a mere technicality. This article therefore seeks to argue that the quasiemployee doctrine should be affirmed in Singapore if or when the opportunity comes before

ibid [69]. Ong Han Ling and another v American International Assurance Co Ltd and others [2018] 5 SLR 549 (”Ong Han Ling”). 21 Munshi Mohammd Faiz v Interpro Construction Pte Ltd and others and another appeal [2021] 4 SLR 1371 (”Munshi”). 22 Ong Han Ling (n 20) [169]–[170]. 23 ibid [175]. 24 Munshi (n 21) [71]. 25 ibid [73]. 26 Ong Han Ling (n 20) [155] and [171]; Munshi (n 21) [68]. 20

295

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

our courts. This article will do so by placing the reasoning and outcome of Bird in the context of the Singaporean legal landscape, and show why Bird is untenable when considered in this context. 7. This article will proceed in the following parts. Part II will explore the reservations the HCA had with accepting the quasi-employee doctrine in Bird, and will show that they can be overcome in the Singapore context. Part III will argue that the quasi-employee doctrine must be accepted in light of the multi-factorial test currently used in Singapore to determine an employment relationship. Part IV will argue that the quasi-employee doctrine is an important tool in the court’s legal toolbox if the doctrine of vicarious liability is to remain a robust tool to achieve the goals of tort law. Part V concludes. II. THE HCA’S RESERVATIONS WITH ACCEPTING THE QUASIEMPLOYEE DOCTRINE CAN BE OVERCOME IN THE SINGAPORE CONTEXT. 8. The HCA in Bird gave three broad reasons for declining to accept the quasi-employee doctrine. First, it rejected the proposition that “[t]he policy objective underlying vicarious liability is to ensure, insofar as it is fair, just and reasonable, that liability for tortious wrong is borne by a defendant with the means to compensate the victim.” 27 The HCA held that this is not a “sound basis for determining and developing the law of vicarious liability and duties of care”28 in Australia. Second, it claimed that some legal propositions (accepted in the UK in Christian Brothers) which were necessary for the quasi-employee doctrine to be accepted (as applied on the Bird fact pattern) were inapplicable in Australia. 29 These legal propositions will be unpacked shortly. Third, it claimed that the expansion of vicarious liability was within the remit of the Legislature rather than the Judiciary. 30 This third point is a matter of institutional competence, which refers to the question of “which branch is best placed due to its expertise and experience [and] its role and function in the constitutional framework of powers” to address that issue.31

Christian Brothers (n 8) [34]. Bird (n 1) [62]. 29 ibid [56]. 30 ibid [63]. 31 Eugene K B Tan, ‘Curial Deference in Singapore Public Law: Autochthonous Evolution to Buttress Good Governance and the Rule of Law’ (2017) 29 SAcLJ 800, para 10. 28

296

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

9. The first reason can be dealt with quickly. The “fair, just and reasonable” basis and the impetus for the victim to be compensated by a defendant with the means to do so, are both present in Singapore, illustrated by the following passage in Skandinaviska v Asia Pacific Breweries:32 “…we are of the view an innocent victim of an employee’s tort should, under ordinary circumstances, be compensated. In this regard, the employer is usually the person best placed and most able to provide effective compensation to the victim. In our view, making the employer vicariously liable is not only a practical solution, but also fair and just.” 33

10. The second and third reasons merit closer attention, which this section aims to do. This section will first argue that the legal propositions rejected in Bird are or should be applicable in Singapore. Second, it will argue that the Singapore courts are in fact institutionally competent to decide on the acceptance of the quasi-employee doctrine. A. The legal propositions in Christian Brothers are or should be accepted in Singapore. 11. In Christian Brothers, four propositions “underpinning the law of vicarious liability in England”34 were identified:35 a. “It is possible for an unincorporated association to be vicariously liable for the tortious acts of one or more of its members” (the “first proposition”). An unincorporated association in Singapore law is a group of individuals who are connected via a contract which is binding amongst its members. 36 b. “D2 may be vicariously liable for the tortious act of D1 even though the act in question constitutes a violation of the duty owed to D2 by D1 and even if the act in question is a criminal offence” (the “second proposition”). c. “Vicarious liability can even extend to liability for a criminal act of sexual assault” (the “third proposition”).

Skandinaviska Enskilda Banken AB (Publ), Singapore Branch v Asia Pacific Breweries (Singapore) Pte Ltd and other appeals [2007] 2 SLR(R) 367 (“Skandinaviska”) 33 ibid [77]. 34 Bird (n 1) [55]. 35 Christian Brothers (n 8) [20]. 36 Chee Hock Keng v Chu Seng Temple [2016] 3 SLR 1396 (“Chee Hock Keng”) [28]–[29].

297

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

d. “It is possible for two different defendants, D2 and D3, each to be vicariously liable for the single tortious act of D1” (the “fourth proposition”). 12. We deal with each of these propositions from least to most controversial. The fourth proposition is largely uncontentious in Singapore, having been accepted in Chen Qiangshi v Hong Fei CDY Construction Pte Ltd. 37 13. The second proposition is slightly more controversial, although the HCA did not comment on its status in Australian law in Bird. It contains two sub-propositions; first, that D2 remains liable even if D1 has breached his duty to D2, and second, that criminal acts generally can cause vicarious liability to attach to an employer. The first sub-proposition cannot be seriously challenged. A tortious act on the part of D1 would almost always be a breach of some kind of a duty to D2 – whether a fiduciary duty or duty of care. Holding that an employee’s breach of duty to their employer would absolve their employer of liability would cause the doctrine of vicarious liability to be almost entirely toothless. 14. On the second sub-proposition of vicarious liability for criminal actions, while there is presently no case in Singapore where an employer was held vicariously liable in tort for an act that was also criminal, it appears that the only barrier to holding so is the rule that the tortious act must be committed in the course of employment. There is no authority which has claims that vicarious liability can never attach simply because the tortious conduct would also attract criminal consequences. The “course of employment” stage is outside the scope of this paper, and thus we say no more on this point, save for a comment related to the third proposition which we will make shortly. 15. The first and third propositions are the most controversial claims. In Bird, the HCA explicitly attacked the first and third propositions, but did not do so for the second and fourth. It stated that the first proposition “does not … reflect the law in Australia”. 38 It further claimed that the third proposition “is in tension with the requirement that the acts must be committed in the course of employment”. 39

Chen Qiangshi v Hong Fei CDY Construction Pte Ltd and another [2014] SGHC 177 (“Chen Qiangshi“) [200]. 38 Bird (n 1) [56]. 39 ibid.

298

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

16. The third proposition goes to the “course of employment” stage of the vicarious liability test, which is, again, outside the scope of this paper. We thus limit ourselves to a passing comment on the HCA’s rejection of the third proposition. We observe that the Singapore CA in Skandinaviska accepted that in the English and Canadian sexual assault cases, the factors in the close connection test were satisfied. 40 It did not appear to have any gripe with the proposition that sexual assault can be committed in the course of employment. We also observe that the close connection test for the “course of employment” stage of the vicarious liability framework,41 finding its roots in Lister v Hesley Hall, had in fact the precise goal of catching acts of sexual assault in its ambit when developed in Lister. This comment also shows that the second proposition is likely to be accepted in Singapore. 17. The first proposition however is a major stumbling block for the quasi-employee doctrine. A significant proportion (although certainly not all) of vicarious liability cases engaging the quasi-employee doctrine concern unincorporated associations. For example, the Diocese in Bird itself was an unincorporated association, as was the Institute in Christian Brothers. 18. The present position in Singapore is that “an unincorporated association … is not a legal entity capable of suing or being sued in its own name” 42 unless it is “clothed with legal personality”43 by statute.44 This creates problems for our position. An unincorporated association which is not registered as a society under the Societies Act thus, prima facie, cannot be vicariously liable for any act of its members simply because it cannot even be named as a defendant in the first place. The position in Chee Hock Keng also does not clarify whether an unincorporated association registered as a society under the Societies Act can be liable for vicarious liability in particular, even if it can be named as a defendant. 19. We thus make the following arguments for how and why unincorporated associations should be vicariously liable for the tortious acts of one or more of its members. First, while we do not disagree that as a general rule, unincorporated associations not registered as societies should not be able to sue or be sued, we will argue that exceptions can be made

Skandinaviska (n 32) [88]. Accepted in Singapore in Skandinaviska (n 32) [75]; applied at [86]–[96]. 42 Chee Hock Keng (n 36) [28]. 43 ibid [28]. 44 One example of such a statutory provision is the Societies Act 1966 (2020 Rev Ed) (the “Societies Act”) s 35(1)(b). 41

299

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

on specific fact patterns where the unincorporated association possesses features which make it similar to a body with legal personality. Second, we will argue that accepting this rule accords with the underlying jurisprudential and policy justifications for the imposition of vicarious liability as a whole. (1) Unincorporated associations should be clothed with legal personality by the court if they possess features which allow them to behave like a single, coherent legal entity. 20. Lord Phillip’s proposition in Christian Brothers that unincorporated associations can attract vicarious liability is admittedly obscure, and has been subject to criticism. The position in the UK on the susceptibility of unincorporated associations is the same as in Singapore. 45 Commentators have pointed out that the cases which Lord Phillips uses to support the first proposition in fact do not go as far as he claims they do. Lord Phillips based his claim on Heaton’s Transport (St Helens) Ltd v Transport and General Workers’ Union [1973] AC 15 (“Heaton’s”), Thomas v NUM (South Wales Area) [1986] Ch 20 (“Thomas”), and Dubai Aluminium Co Ltd v Salaam [2003] 2 AC 366 (“Dubai Aluminium”).46 But, as Professor Claire McIvor points out, in Dubai Aluminium, the “unincorporated association” there was a partnership, which is clothed with legal personality in statute, specifically s 10 of the UK’s Partnership Act 1890.47 Further, it is unclear how Heaton’s and Thomas support the first proposition at all – as Professor Phillip Morgan points out, those cases appear to concern the liability of a principal for its agent’s torts, rather than vicarious liability.48 Thus, it appears that Christian Brothers has, as a novel development of law, carved out a somewhat strange exception to the general proposition that an unincorporated association not clothed with legal personality by statute cannot sue or be sued. 21. But this exception is not as inexplicable as it seems. Professor McIvor points out that the real roots of this exception arises from the English Court of Appeal (“EWCA”) case of E v English Province of Our Lady of Charity (“JGE”).49 There, the EWCA held the Catholic

See Aireborough Neighbourhood Development Forum v Leeds City Council [2020] 1 WLR 2355 [13]; London Association for Protection of Trade v Greenlands Ltd [1916] 2 AC 15 [33]. 46 Christian Brothers (n 8) [20]. 47 Claire McIvor, ‘Vicarious liability and child abuse’ (2013) 29(1) PN 62, 64. See also Phillip Morgan, ‘Vicarious liability on the move’ (2013) 129 LQR 139, 141. 48 Morgan, ‘Vicarious liability on the move’ (n 47) 141. 49 [2013] QB 722. See McIvor (n 47), 64. The case is cited in McIvor’s article as JGE v The Trustees of the Portsmouth Roman Catholic Diocesan Trust – the different names refer to the same case. The case will be referred to as JGE to avoid confusion because the case was referred to as such in Christian Brothers, which we quote shortly.

300

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Church vicariously liable for acts of sexual abuse committed by a priest. 50 The Catholic Church was “treat[ed] exactly like a business”51 and held vicariously liable because it possessed features which made it analogous to a business.52 It was “highly organised and extremely wealthy”,53 had great “power and influence”, 54 and “operate[d] in a highly organised and hierarchical fashion”. 55 In Christian Brothers, the UKSC decided that “the case for finding vicarious liability [was] much stronger in the present case than it was in JGE” 56 and thus, by analogy, imposed liability on the Institute. 22. What appears to have happened is that the court clothed the Catholic Church in JGE and the Institute in Christian Brothers with legal personality despite there being no statutory authority for it, because the Catholic Church and the Institute exhibited features which made them appear to be a single, coherent legal entity (viz, the high degree of organisation and the significant degree of “power and influence”). It was fair to regard these organisations as having legal personality because of these features. This is thus not an arbitrary exception to the general rule that an unincorporated association without statutorily granted legal personality cannot sue or be sued. It is instead a fact-sensitive approach that requires specific features to be met before the court clothes the organisation with legal personality. 23. The approach here, it is submitted, is a principled means of allowing an unincorporated association to attract legal consequences where their members commit torts. It is thus submitted that the Singapore courts should follow this fact-sensitive approach to clothing unincorporated associations with legal personality to enable them to be vicariously liable for a tort of a quasi-employee. 24. A question might arise as to whether this approach can be utilised outside of vicarious liability to hold an unincorporated association directly liable in tort (or any other reason for legal liability for that matter). This point is outside the scope of this article, but we make some brief observations. In Meridian Global Funds Management Asia Ltd v Securities

JGE at [83]–[84]. McIvor (n 47) 64. 52 JGE (n 49) [77]. 53 McIvor (n 47) 64. 54 ibid. 55 ibid. 56 Christian Brothers (n 8) [55]. 51

301

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Commission,57 the Privy Council identified three types of rules which allow the acts of a natural person to be attributed to the company that natural person is acting on behalf of: primary, general and special rules. 58 These rules of attribution were accepted in Singapore in Ho Kang Peng v Scintronix Corp 59 for the purpose of attributing civil liability to a company.60 Vicarious liability is one of the general rules of attribution. 61 There appears to be no reason, in principle, why the other types of rules cannot be used to directly attribute liability to an unincorporated association if that association functions in effect like a company. 25. Finally, it is pointed out that even if this point is not accepted, it will not be fatal to the applicability of the quasi-employee doctrine generally. The arguments that follow will still apply to members of unincorporated associations which are statutorily clothed with legal personality, as well as for any other associations with legal personality which engage potential tortfeasors on a quasi-employee basis. One such example is the case of Cox v Ministry of Justice.62 There, a prisoner who was assigned a job in the prison kitchen negligently dropped a sack of rice onto the claimant’s back, causing injury to the claimant. 63 The Ministry of Justice was held vicariously liable for the prisoner’s negligence. 64 The quasi-employee doctrine was a necessary part of the UKSC’s reasoning.65 Thus, the quasiemployee doctrine can apply even outside an unincorporated association context, making the first proposition not as large a stumbling block as the HCA makes it out to be. (2) Allowing vicarious liability to attach to unincorporated associations accords with the underlying jurisprudential and policy justifications for the imposition of vicarious liability. 26. In Skandinaviska, it was accepted that there were two policy considerations underlying the doctrine of vicarious liability: first, “effective compensation for the victim” and second, “deterrence of future harm by encouraging the employer to take steps to reduce risk of

Meridian Global Funds Management Asia Ltd v Securities Commission [1995] 2 AC 500 (“Meridian”). ibid 507. 59 Ho Kang Peng v Scintronix Corp Ltd (formerly known as TTL Holdings Ltd) [2014] 3 SLR 329 (“Scintronix”). 60 ibid [48]. 61 Meridian (n 57) 507; Scintronix (n 59) [48]. 62 Cox v Ministry of Justice [2016] AC 660 (“Cox”). 63 ibid [5]. 64 ibid [32]. 65 ibid [27]. 58

302

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

similar harm in future”.66 This section aims to show that vicarious liability for unincorporated associations achieves these policy objectives. 27. Some observations are apt before we begin our argument proper. First, it is accepted that “effective compensation for the victim” is not a trump card that justifies an employer having to pay for their employee’s tort in every instance. This is firstly a matter of fairness to the employer, for it would be intuitively unfair for the employer to be liable for something which it had absolutely nothing to do with. It is also a matter of giving credence to the second goal of deterrence, for there would be nothing to deter if the employer had no control over the employee’s action, and the only situation in which the employer can control the employee is if that employee’s action was committed in the course of his job. This rationale explains the rule that the tortfeasor’s act must have a “close connection” with the employer’s business. 28. That having been said, allowing vicarious liability to attach to unincorporated associations generally (whether registered as a society or not) is in line with the goal of effective compensation for the claimant, without compromising fairness to the association. Vicarious liability is an appealing route for a claimant where it is available because suing the tortfeasor’s institution is often the most promising option for compensation. This is because the institution is likely to have more means to compensate the claimant as opposed to a single tortfeasor. Further, in some instances, perhaps the individual tortfeasor cannot be identified, or is dead (as was the case in Bird).67 In such situations the best, or possibly only, way the claimant can seek redress is to sue the tortfeasor’s organisation. 29. It might be said that this rationale should not apply to an unincorporated association (whether registered as a society or not). This would be true if the reasons for vicarious liability to attach to an ordinary defendant do not apply to unincorporated associations. We now assess this claim.

Skandinaviska (n 32) [76]. See also Paula Giliker, ‘Analysing institutional liability for child sexual abuse in England and Wales and Australia: vicarious liability, non-delegable duties and statutory intervention’ (2018) 77 CLJ 506, 511. 67

303

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

30. In Christian Brothers, Lord Phillips laid down the following reasons which justify vicarious liability attaching to an institution, 68 cited favourably by the CA in Ng Huat Seng:69 a. “The employer is more likely to have the means to compensate the victim than the employee …”. b. “The tort will have been committed as a result of activity being taken by the employee on behalf of the employer.” c. “The employee’s activity is likely to be part of the business activity of the employer.” d. “The employer, by employing the employee to carry on the activity will have created the risk of the tort committed by the employee.” e. The employee will, to a greater or lesser degree, have been under the control of the employer. 31. Unincorporated associations do not necessarily lack in any one of these five features. Whether or not the unincorporated association lacks any one of these features, is a question that turns on the facts of each case. As such, it is not uniformly unfair to impose liability on an unincorporated association in all cases 32. On the first feature, an unincorporated association could have a common pool of funds, from which damages can be drawn from, or alternatively, the damages could simply be spread pari passu across the association’s membership. Thus, it is entirely possible for an unincorporated association to have greater means than the tortfeasor to compensate a victim. 33. On the second to fourth features, an unincorporated association would certainly have been formed for a particular purpose, discernible on the facts of each case. This is similar to a traditional company, whose “business activities” are determined on an analysis of the facts. Thus, the task of parsing out the supposed “activity” of an unincorporated association is

68 69

Christian Brothers (n 8) [35]. Ng Huat Seng (n 18) [54].

304

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

not much different from parsing out that of a traditional company. And it must have people who carry out tasks to further that purpose, just like a traditional company. 34. On the fifth feature, it is admitted that an unincorporated association would likely have looser control over its members than a traditional company. Nonetheless, it is entirely reasonable for an unincorporated association to be considered to have some degree of control over its members, since its members are bound by contractual terms which are set by the leaders of the association. Leaders in these unincorporated association also can set internal rules to govern their members. Further, modern companies too do not have great degrees of control over their employees; 70 yet, this has not stopped courts from imposing vicarious liability on traditional companies. 35. Thus, all five reasons for imposing vicarious liability on traditional companies can also apply to unincorporated associations. B. The Singapore courts are institutionally competent to decide on the expansion of vicarious liability 36. In Bird, the HCA stated that “[i]n light of this Court having rejected, on more than one occasion over the last 25 years, both the starting point and the basis on which the Supreme Court of the United Kingdom extended the law of vicarious liability, the issue is squarely in the hands of the legislatures.”71 It appears that the HCA found itself lacking institutional competence to rule on the issue, finding Parliament the better forum for the matter instead. 37. It is submitted that this reasoning does not apply to Singapore. First, as explained above, the starting points and rationale underlying the extensions of vicarious liability created by the UKSC are applicable in Singapore. Second, the courts are institutionally competent to decide on the expansion of vicarious liability because Singapore’s Parliament has never discussed negligence and vicarious liability, unlike Australia’s federal and state Parliaments which have done so. This section will focus on the second point. 38. The court’s institutional competence lies in applying the law to specific cases, as well as developing the “principles of common law and equity”.72 Its powers of law making in the 70

Christian Brothers (n 8) [36]. Bird (n 1) [63]. 72 Lim Meng Suang and another v Attorney-General and another appeal and another matter [2015] 1 SLR 26 (“Lim Meng Suang”) [77]. 71

305

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

form of developing the common law are subservient only to Parliament where Parliament decides to intervene via legislative law-making.73 Conversely, if Parliament has considered a change to the law and decided not to enact that change, the courts will generally decline to make a similar change in the common law. 74 39. Australia’s federal Parliament commissioned a Review of the Law of Negligence in 2002, 75 and various state legislatures enacted statutory reforms to the law of negligence as a result. In relation to vicarious liability, the Ipp Report panel considered whether community organisations should be vicariously liable for torts committed by volunteers. It recommended against expanding vicarious liability in such a manner. 76 This would be a similar expansion of vicarious liability to accepting the quasi-employee doctrine, in the sense that it is an expansion of the scope of vicarious liability to accept relationships outside of a traditional employment context. Further, as observed by the majority in Bird,77 state level legislatures have also considered reforms to vicarious liability and have done so in differing ways – the New South Wales state Parliament, for example, modified the state’s Civil Liability Act to statutorily enshrine the quasi-employee doctrine,78 while some other states have not. It is thus understandable that the Australian courts should be hesitant to widen the ambit of the employment stage of the vicarious liability test, when federal and state level legislatures have already considered the matter, and state level legislatures have come to differing conclusions on the issue. The courts understandably do not want to usurp what should be within the proper remit of the Legislative branch, both at the federal and state level. 40. Unlike in Australia, Singapore’s Parliament has never considered reforms to the law of vicarious liability. Vicarious liability thus remains firmly in the hands of the common law. And indeed, the Singapore courts have consistently regarded the development of vicarious liability as within their judicial remit. This is evident both from the obiter dicta in Ng Huat Seng on the employment stage of the vicarious liability test, as well as the acceptance of

See Lim Meng Suang (n 72) [79]. See also Goh Yihan, ‘Where judicial and legislative powers conflict: Dealing with legislative gaps (and non-gaps) in Singapore’ (2016) 28 SAcLJ 472 482–483. 74 See eg Yong Vui Kong v Public Prosecutor and another matter [2010] 3 SLR 489 [92], where the CA declined to interpret a prohibition against inhuman punishment into the Constitution on the basis that Parliament had explicitly rejected the 1966 Constitutional Commission’s recommendation to add such a prohibition. 75 Review of the Law of Negligence (Final Report, September 2002) (“Ipp Report”). 76 ibid [11.20]–[11.23]. 77 Bird (n 1) [68]. 78 Civil Liability Act 2002 (NSW) s 6G.

306

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

the close connection test for the “course of employment” stage in Skandinaviska, both of which have been referred to earlier in this paper. 41. Accordingly, the legal landscape in Australia and Singapore differ greatly with regard to the court’s institutional competence to expand the law of vicarious liability. If the opportunity arises to recognize the quasi-employee doctrine in Singapore, institutional competence should be no barrier to the courts doing so. III. THE MULTI-FACTORIAL TEST CURRENTLY USED IN SINGAPORE TO DETERMINE AN EMPLOYMENT RELATIONSHIP RENDERS REJECTION OF THE QUASI-EMPLOYEE DOCTRINE LOGICALLY FLAWED A. The formalistic distinction adopted in Bird is inconsistent with Singapore's multi-factorial test for employment. 42. In Singapore, the current test for identifying a qualifying employment relationship in the context of vicarious liability is multifactorial, 79 encompassing considerations of control, integration, and economic reality. 80 This was affirmed in BNM, where the CA criticized the lower court’s exclusive reliance on the “control” test, 81 holding that other factors – such as the degree of personal investment and financial risk undertaken in carrying on a business – are “equally material” in distinguishing an employee from an independent contractor. 82 43. Historically, the existence of an employment relationship turned almost conclusively on the extent of control one party exercised over the other's manner of work. 83 This reflected the medieval origins of vicarious liability, under which a “master” was held responsible for wrongs committed by his “servants” in the course of their duties precisely because of the extensive control he exercised over their performance.84

BNM (administratrix of the estate of B, deceased) on her own behalf and on behalf of others v National University of Singapore and others and another appeal [2014] 4 SLR 931 (“BNM”) [31]–[33]; Low Kee Yang, ‘Vicarious liability, non-delegable duty and the Ng Huat Seng decision’(Singapore Law Gazette, December 2017) < https://lawgazette.com.sg/feature/vicarious-liability-non-delegable-duty-ng-huat-seng-decision/> accessed 3 July 2026. 80 Low (n 79). 81 BNM (n 79) [31]–[32]. 82 ibid [32]. 83 MCST No 3322 v Mer Vue Developments Pte Ltd and others [2016] 2 SLR 793 [11]. 84 Warren Swain, ‘A historical examination of vicarious liability: A ’veritable upas tree’?’ (2019) 78(3) CLJ 640, 650.

307

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

44. However, as notions of worker autonomy have come to underpin modern organizational ethics, and professional and highly skilled work has become increasingly prevalent, the degree of control employers exercise over the manner of work of their employees has diminished significantly.85 Consequently, control alone is no longer sufficient in discerning the existence of an employment relationship. This therefore prompted the courts in BNM and Market Investigations to call for a more holistic, multifactorial approach to identifying such relationships.86 Indeed, as Cooke J observed in Market Investigations, cited with approval in BNM, the shifting dynamics of modern employment means that “the absence of control and direction...can be little, if any use as a [decisive] test.” 87 This recognition gave impetus to the multifactorial approach now prevailing in both jurisdictions. 45. Further, the court in BNM endorsed Denning L.J.’s observation in Market Investigations that “[n]o exhaustive list has been compiled and perhaps no exhaustive list can be compiled of the considerations which are relevant in determining [whether an employment relationship exists], nor can strict rules be laid down as to the relative weight which the various considerations should carry in particular cases.”88 46. The acknowledged difficulty in identifying the relevant factors and assigning them appropriate weight reflects the fact that employment relationships manifest in many different forms and are continually evolving. A rigid, single test is therefore inadequate to capture their full and evolving diversity. 89 This was recognized by the Court of Appeal CA in Ng Huat Seng, which held that the doctrine of vicarious liability must evolve to “accommodate the more diverse range of relationships which might be encountered in today's context,” – relationships that nonetheless possess the same fundamental qualities as traditional employment that justified the imposition of vicarious liability in the first place. 90 47. It follows that denying vicarious liability simply because a relationship does not conform to "traditional" employment is inconsistent with Singapore's accepted premise that employment relationships are inherently “diverse”. Yet, this was precisely the basis on

Market Investigations Ltd v Minister of Social Security [1969] 2 QB 173 (“Market Investigations”) 183–185. BNM (n 79) [31]; Market Investigations (n 85) 183–185. 87 Market Investigations (n 85) 183. 88 ibid 184–185. 89 Douglas Freeburn, ‘Employment or nothing’: An Analysis of Bird v DP and the Australian Approach to Vicarious Liability in Cases of Institutional Abuse (2025) 44(1) University of Queensland Law Journal 183, 201. 90 Ng Huat Seng (n 18) [63]. 86

308

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

which the court in Bird rejected the quasi-employee doctrine. 91 Such a reasoning is untenable in Singapore's context and would allow mere formalistic distinctions to override the policy justifications for imposing vicarious liability that the Court in Ng Huat Seng identified as central to determining which relationships are capable of founding the doctrine.92 48. Hence, we respectfully argue that the Singaporean courts should decline to follow Bird in future cases. B. Dismissing the quasi-employee doctrine on the ground that it is not a “strict” employment relationship is arbitrary at best and circular at worst. 49. In Bird, the majority held that a strict relationship of employment must exist between the tortfeasor and the defendant before vicarious liability can arise. 93 However, the majority failed to explain why no such relationship existed between Coffey and the Diocese, abruptly concluding that their relationship was “sui generis” and therefore incapable of satisfying the threshold requirement of a qualifying employment relationship. 94 This failure to reason through the threshold question exposes the logical flaws underlying the majority's dismissal of the claim. 50. In concluding that no “strict” employment relationship existed between Coffey and the Diocese, the majority must have proceeded on the assumption that there exists, in the first place, a determinate conception of “strict” employment which the relationship between the Diocese and Coffey failed to satisfy. 95 51. This assumption, however, runs counter to Singapore’s multifactorial test for determining employment. As explained in part III (A) of this article, that test proceeds on the recognition that employment is too elusive and dynamic a concept to admit of a single, intuitive definition. That premise is fundamentally inconsistent with the HCA’s notion that “strict” employment can be coherently conceived, rendering the majority’s decision in Bird incoherent within Singapore's framework.

Bird (n 1) [47]. Ng Huat Seng (n 18) [63]. 93 Bird (n 1) [47]. 94 ibid [234]. 95 ibid [8]. 92

309

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

52. Moreover, with the meaning of “strict” employment left wholly undefined and unelaborated by the majority in Bird (precisely because it is nigh impossible to admit of fixed defining characteristic(s)), the decision in Bird is arbitrary at best and circular at worst. To conclude that no qualifying employment relationship exists based on the absence of a relationship satisfying an undefined notion of “strict” employment renders the court in Bird complicit in precisely the uncertainty and indeterminacy it sought to avoid in rejecting the quasi-employee doctrine. 96 In fact, the majority’s approach arguably creates greater uncertainty than the quasi-employee doctrine itself. Whereas the former determines the existence of a qualifying employment relationship through an undefined notion of “strict” employment, the latter does so through a multifactorial analysis, analogous in substance and rigour97 to that used in identifying relationships of employment. 53. At its worst, the conclusion that no qualifying employment relationship exists because there is no “strict” employment, absent any definition of the latter, amounts to circular reasoning. We respectfully submit that future Singapore courts ought not to rely on a decision that falls short of acceptable legal reasoning. IV. THE QUASI-EMPLOYEE DOCTRINE ENABLES VICARIOUS LIABILITY TO CONTINUE TO BE A ROBUST TOOL WHICH ACHIEVES THE GOALS OF TORT LAW. 54. We now offer some positive arguments for the acceptance of the quasi-employee doctrine. We propose two reasons why the quasi-employee doctrine is an important part of the tort law landscape in Singapore. First, the quasi-employee doctrine allows the law to keep up with the new relationships replacing traditional employment in the modern commercial context. Second, the quasi-employee doctrine fulfils the goals of tort law in a fair and calibrated manner. A. The quasi-employee doctrine allows the law to keep up with “the more diverse range of relationships which might be encountered in today’s context.”98 55. The modern business world features a wide range of complex relationships other than that between the employer and employee. As Professor David Tan points out, “businesses in

ibid [241]. Freeburn (n 89) 201. 98 Ng Huat Seng (n 18) [63]. 97

310

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

the twenty-first century are increasingly relying on call centres, online third-party sellers, sales agents, brokers, ride-sharing contract drivers, debt collectors, delivery services and many other external providers that may harm third parties.” 99 These new relationships represent a shift from the traditional model of employment, potentially causing problems for the application of the employment stage of the vicarious liability framework. This section will focus on two changes to the landscape of employment today – the gig economy and artificial intelligence (“AI”) – and argues that the quasi-employee doctrine is an important legal tool to ensure the continued relevance of the doctrine of vicarious liability. (1) The quasi-employee doctrine helps vicarious liability to stay relevant in the gig economy. 56. The gig economy refers to the phenomenon where employment is characterised by workers “tak[ing] up short jobs … on an ad hoc basis and earn[ing] a fee for performing this work”,100 typically paid “in accordance with the specific amount of work required to be performed”101 rather than with reference to a specific period of time or by a complete piece of work. 57. The gig economy poses a problem for the employment stage of the vicarious liability framework because both the appearance and the substance of the relationship between a gig worker and their employer differs. On appearance, “gig economy enterprises go to lengths to avoid structures that resemble contracts of service, or terminology that suggests employment”.102 And on substance, while in a traditional employer-employee relationship obliges the employer “provide the worker with work” and obliges the employee to “accept this work”,103 the same obligations do not exist in a gig worker relationship for both employer and employee. 104 It is also noted that gig workers are now statutorily recognised as distinct from traditional employees in Singapore. Gig workers are given special statutory

David Tan, ‘Enterprise risk and vicarious liability in Singapore’, in Paula Giliker (ed), Vicarious Liability in the Common Law World (Bloomsbury 2022) 158. 100 Amy Seow and Alina Chia, ‘Vicarious liability and enterprise risk in the gig economy’ [2020] SAL Prac 1, para 2. 101 ibid. 102 ibid, para 5. 103 ibid, para 4. 104 ibid, paras 4–5.

311

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

classification by the Platform Workers’ Act 2024, 105 which is said to “in effect [create] a middle category of workers in between employees and self-employed persons”.106 58. This means that gig workers are very likely to fall outside of the category of employee in the traditional sense. Thus, vicarious liability may not attach to employers of gig workers who commit torts if insistence on a traditional relationship of employment is demanded. Yet at least one commentator has suggested that the presently accepted underlying rationale of vicarious liability in Singapore of “enterprise risk” 107 calls for the outcome that employers attract vicarious liability for a gig worker’s torts.108 Any debate on whether this is correct is outside the scope of this article. We stop at noting that if vicarious liability is to be extended to torts committed by gig workers, the quasi-employee doctrine must be accepted given that gig workers both contractually and statutorily do not fall within the category of a traditional employment relationship. 59. It might be argued that Australia successfully dealt with the new relationships created by the gig economy without the quasi-employee doctrine in Hollis v Vabu.109 There, an unidentified cyclist who was working as a courier for a company called Vabu negligently crashed into one Mr Hollis, causing injury to Mr Hollis. Mr Hollis sued Vabu, arguing that it was vicariously liable for the courier’s negligence. The contract between Vabu’s couriers and Vabu were described as “contract[s] for service”110 – in other words, a contract with an independent contractor, 111 as is common in the gig economy as explained earlier. Nonetheless, the HCA held that the unknown cyclist was in a relationship of employment with Vabu and held Vabu vicariously liable. 112

105

Platform Workers Act 2024 s 5. Singapore Parliamentary Debates, Official Report (9 September 2024), vol 95 (Dr Koh Poh Koon, Senior Minister of State for Manpower). 107 The enterprise risk rationale suggests that vicarious liability is imposed on an employer because “carrying on a business enterprise necessarily involves risk to others … when those risks ripen into loss, it is just that the business should be responsible for compensating the person who has been wronged”: Dubai Aluminium Co Ltd v Salaam [2003] 2 AC 366 [21], cited in Seow and Chia (n 100), para 27. 108 Tan (n 99). 109 Hollis v Vabu Pty Limited (2001) 207 CLR 21 (“Hollis”). 110 ibid [23]. 111 Gary Chan Kok Yew and Lee Pey Woan, The Law of Torts in Singapore (2nd edn, Academy Publishing 2016) para 19.009. 112 Hollis (n 109) [47]. 106

312

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

60. However, Hollis has been characterised by commentators as a case which “acknowledge[s] that vicarious liability can extend beyond situations of strict employment”. 113 Indeed, the HCA in Hollis appeared to apply a multi-factorial test similar to that of the quasi-employee doctrine, rather than the formalistic “strict” employment test in Bird. In Hollis, the HCA identified an employment relationship between the courier and Vabu by reference to the latter's power to control the performance of the work114 and the degree to which the courier was integrated into the business,115 rather than by appealing to any formalistic conception of “strict” employment. 61. We venture further and posit that the HCA in Hollis in fact applied the quasi-employee doctrine.

The

quasi-employee

doctrine

said

“supplement” 116

the

employee/independent contractor dichotomy, “extending vicarious liability to various sui generis relationships which [the dichotomy] does not encompass”. 117 Now consider the following passage in McHugh J’s judgment in Hollis, who concurred with the majority that Vabu was vicariously liable (albeit with slightly different reasoning): Rather than expanding the definition of employee or accepting the employee/independent contractor dichotomy, the preferable course is to hold that employers can be vicariously liable for the tortious conduct of agents who are neither employees nor independent contractors … [this] has the great advantage of ensuring that the doctrine of vicarious liability remains relevant in a world of rapidly changing work practices.118

62. As can be seen, what the court intended to do in Hollis and what the quasi-employee test sets out to do are one and the same. They seek to look beyond the form of the contract the tortfeasor is under to render the organisation vicariously liable despite the lack of an explicit relationship of employment. What this means for our purposes is that Hollis does not represent a successful handling of a gig relationship despite the lack of the quasi-employee doctrine. It is instead a representation that the quasi-employee doctrine is necessary to handle the relationships brought about by the gig economy.

113

Anthony Gray, ‘Relationships akin to employment and vicarious liability: The High Court’s rejection in Bird v DP (A Pseudonym)’ (2025) 29 Tort Law Review 376 (“Gray”), 386. 114 Hollis (n 109) [49]. 115 ibid [50] and [102]. 116 Donal Nolan and Ken Oliphant, Lunney & Oliphant’s Tort Law: Text and Materials (7th edn, Oxford University Press 2023) 842. 117 ibid. 118 Hollis (n 109) [93]. See also Gray (n 113) 385–386, where Professor Gray discusses a similar point and cites the same passage in Hollis (n 109) at footnote 71.

313

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

(2) The quasi-employee doctrine helps vicarious liability stay relevant in the world of AI. 63. Second, and perhaps more pressingly, AI is now increasingly being integrated into the commercial landscape (whether we like it or not). AI potentially causes problems for the doctrine of vicarious liability because AI is increasingly used to replace traditional human employees, so much so that the doctrine of vicarious liability has been said to be at risk of being “obsolete”.119 64. The quasi-employee doctrine however “offers the possibility of the doctrine [of vicarious liability] applying to the torts of non-natural persons”,120 particularly AI. While there has yet to be a case imposing vicarious liability for a tort committed by AI, absent the quasiemployee doctrine, there appears to be no other legal tool by which vicarious liability may attach to a tort committed by AI. Accepting the quasi-employee doctrine thus offers a potential lifeline to the doctrine of vicarious liability in a world where human employees are increasingly being replaced by AI tools. B. It is in the interests of the goals of tort law to accept the quasi-employee doctrine. 65. The Singapore courts have recognised at least three main goals of tort law: compensation (or corrective justice),121 distribution of risk (or distributive justice), 122 and deterrence.123 Accepting the quasi-employee doctrine would fulfil all three of these goals. 66. First, the quasi-employee doctrine fulfils tort law’s compensatory goal because without it, a victim of a tort committed by an individual who does not fulfil a “strict” relationship of employment may be left without an adequate remedy. As explained above in the discussion on unincorporated associations, where vicarious liability is an option, often it would be easier for the claimant to seek a remedy from the institution rather than the tortfeasor themselves.

119

Phillip Morgan, ‘Tort law and AI’, in Ernest Lim and Philip Morgan (eds), ‘The Cambridge Handbook of Private Law and Artificial Intelligence’ (Cambridge University Press, 2024) 138–139. 120 ibid 139. 121 Tan Juay Pah v Kimly Construction Pte Ltd and others [2012] 2 SLR 549 [84]. 122 ibid [85]; Spandeck Engineering (S) Pte Ltd v Defence Science & Technology Agency [2007] 4 SLR(R) 100 [29]. 123 Salmizan bin Abdullah v Crapper, Ian Anthony [2024] 5 SLR 257 [48].

314

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

67. At the same time, the quasi-employee doctrine is not a rule that expands liability in such a radical manner that organisations are unfairly burdened with legal risk. The quasi-employee doctrine has mechanisms to ensure that liability is only imposed where the facts make it fair to do so. This comes in the form of the multi-factorial framework to determine whether a quasi-employee relationship in fact appears on the facts. And, of course, the ‘course of employment’ stage of the vicarious liability framework adds an added layer of protection for the organisation. Thus the quasi-employee doctrine is a controlled method of achieving corrective justice which does not expand liability unfairly 68. Second, vicarious liability also serves a distributive function, which acceptance of the quasi-employee doctrine would enhance. By making the institution liable for the tortfeasor’s conduct, the consequences of a tort being committed is redistributed amongst the members of the institution.. This distribution is fair because the organisation creates the risk of the tort occurring to the claimant as a result of setting the tortfeasor to work for the organisation’s benefit. It is only fair that that risk is redistributed from the claimant to the organisation, or else the organisation gets all the benefit of the tortfeasor’s work, yet bears none of the associated risk. 69. The quasi-employee doctrine is a natural extension of this logic. A quasi-employee acts in furtherance of the goals of the organisation they work for. It is only right that the organisation bears the risk that is borne out of such activity which the organisation benefits from. There is no reason why a relationship which falls outside a “strict” employment relationship should not be subject to this same logic. Rejection of the quasi-employee doctrine would foist the risk and consequences of the tort upon the claimant, a clearly unfair distribution for the claimant who bears the entirety of the risks arising from the activities, while the organisation simultaneously reaps the rewards of those very activities. 70. Third, the quasi-employee doctrine enhances the deterrent effect of tort law. Vicarious liability in particular aims to encourage organisations to keep tight control over their employees, and to deter an overly ‘laissez-faire’ management style which may increase the risk of employees harming others in the course of their employment. It is instructive that control is a key factor in the multi-factorial framework for finding a relationship of employment. Thus, cases which engage the quasi-employee doctrine would necessarily feature a significant degree of control by the organisation over the tortfeasor. Indeed, in

315

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Bird, the Diocese exercised “at least as great, if not greater” 124 control over the terms of the priest’s appointment as would typically apply to an employee. If the quasi-employee doctrine were not accepted, there would be no incentive for organisations to control the individuals it assigns to do work that benefits the organisation, even though the levers of control are present. The quasi-employee doctrine thus serves as a spur for organisations to exercise control over anyone who works for the benefit of the company, if they have the means to do so. V. CONCLUSION 71. This paper has argued that the quasi-employee doctrine should be accepted by the Singapore courts despite its rejection in Australia for the following reasons. First, the reservations the HCA had with accepting the quasi-employee doctrine in Bird can be overcome in the Singapore context. Second, the quasi-employee doctrine must be accepted in light of the multi-factorial test currently used in Singapore to determine an employment relationship. Third, the quasi-employee doctrine enables vicarious liability to continue to be a robust tool which achieves the goals of tort law. 72. The quasi-employee doctrine has proven itself to be a useful legal tool in several cases in the UK to allow the court to provide relief to a deserving claimant. 125 We are confident that it will similarly prove its value in the Singapore context. It is hoped that this article aids both the court and litigants if the outcome in Bird is ever argued for before the Singapore court.

124

Bird (n 1) [167], citing the lower court judgment: Bird v DP (a pseudonym) (2023) 69 VR 408 [125]. Apart from Christian Brothers (n 8) and Cox (n 62), other significant examples include Armes v Nottinghamshire County Council [2018] AC 355 and Trustees of the Barry Congregation of Jehovah’s Witnesses v BXB [2024] AC 567. 125

316

STEP INSIDE YOUR TSMP CAREER WHO WE ARE At TSMP Law Corporation, we hire future partners, not faceless associates. We’re Singapore’s pre-eminent independent corporate boutique, trusted on complex disputes and cutting-edge deals. WHAT WE’RE LIKE Grades and a burnished CV may get you in the door, but it is your strength of character, personality and partnership potential that get you a place at the table. Our teams are nimble, our culture collaborative and our office a creative cove buzzing with conversation, banter and big ideas. WHY WE DO WHAT WE DO Above all, we are sharks with heart. Every lawyer serves 25 hours of pro bono work a year, and we shut the firm for two full days annually to do charity work. Diversity and inclusivity are at our core: we support SG Her Empowerment in its mission to fight online harms against women, and we believe that serving the community is a privilege, not a tagline. HOW YOU’LL GROW As a junior lawyer, you don’t sit on the sidelines. From day one, you have frequent touch-points with senior lawyers and Partners, and you’re thrown into high-level corporate and dispute resolution work throughout your training. You get real responsibility, real clients and real impact.

START YOUR CAREER JOURNEY WITH US

www.tsmplaw.com @tsmplaw TSMP Law Corporation

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

“HOW IS THIS LEGAL?”: REASSESSING MULTI-LEVEL MARKETING REGULATION IN SINGAPORE Edison Toh* ABSTRACT Singapore prohibits pyramid selling schemes under the Multi-Level Marketing and Pyramid Selling (Prohibition) Act 1973, but permits qualifying multi-level marketing (“MLM”) arrangements under the Multi-Level Marketing and Pyramid Selling (Excluded Schemes and Arrangements) Order to operate. While this framework formally distinguishes lawful MLMs from unlawful pyramid schemes, persistent concerns remain regarding participants’ monetary losses and exaggerated product claims. These developments raise the question: To what extent does Singapore’s regulatory framework effectively distinguish lawful MLM schemes from unlawful pyramid schemes, and does it provide meaningful protection for participants against financial and psychological harm? This article examines whether Singapore’s statutory safeguards adequately distinguish legitimate direct-selling businesses from product-based pyramid schemes (“PPS”). This article analyses Singapore’s regulatory framework alongside the U.S. approach developed under § 5 of the Federal Trade Commission Act, reflected in FTC enforcement and key judicial decisions. It argues that the principal difference between the two jurisdictions is in the legal inquiry adopted to distinguish lawful businesses from unlawful pyramid schemes. While Singapore primarily relies upon compliance with prescribed statutory safeguards and the formal attribution of benefits to product sales, the U.S. adopts a substance-based assessment that examines whether compensation is economically sustained by genuine retail demand or by recruitment-driven incentives. The article submits that Singapore’s existing framework permits sophisticated MLM models to circumvent the safeguards of the Exclusion Order. It therefore recommends supplementing the current framework with an economic reality assessment, introducing mandatory disclosures of distributor earnings, and considering a licensing and competency framework for MLM promoters, modelled on other regulated industries. These reforms would better align Singapore’s regulatory framework with Parliament’s original *

University of Bristol, LLB (Hons), Class of 2027. I am grateful to the editorial reviewers for their thoughtful comments on earlier drafts. Any remaining errors are my own. E-mail for correspondence: edisontoh.bh@gmail.com

317

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

objective of protecting consumers from the substantive harms associated with pyramid selling while preserving legitimate direct-selling enterprises. I.

INTRODUCTION

Despite existing statutory safeguards that distinguish lawful Multi-Level Marketing (“MLM”) schemes from unlawful pyramid schemes and ultimately protect participants, concerns about participant harm remain significant. 1 Broadly speaking, MLMs are product-based direct-selling businesses in which participants may earn compensation through multiple levels of distribution,2 whereas pyramid schemes primarily reward participants for expanding the participant network itself. 3 While the two schemes appear structurally different, the distinction between them is often less clear than it first appears. Empirical data report that, on average, 99% of MLM participants incur net financial losses after accounting for product purchases and business-related expenses.4 CNA Insider’s Talking Point’s 5 investigation into MLM schemes further highlights issues such as exaggerated product claims 6 and misleading representations of earning potential. 7 Beyond financial harm, psychological research identifies recurring concerns relating to emotional manipulation, social coercion and the exploitation of vulnerable individuals.8 More recently publicised is the murder of 19-year-old Ms Huang Baoying, which 1

CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 2:44 (‘Why MLMs have a bad reputation’). 2 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 1:16 (‘How do MLMs work?’). 3 Federal Trade Commission, Business Guidance Concerning Multi-Level Marketing (26 April 2024) subheading ‘What is multi-level marketing?’ https://www.ftc.gov/business-guidance/resources/businessguidance-concerning-multi-level-marketing#_ftnref1 4 Jon M Taylor, The Case (for and) Against Multi-Level Marketing (submitted to the Federal Trade Commission, 2008) appendix 7A https://www.ftc.gov/sites/default/files/documents/public_comments/traderegulation-rule-disclosure-requirements-and-prohibitions-concerning-business-opportunities-ftc.r51199300008%C2%A0/00008-57281.pdf accessed 15 June 2026 5 Talking Point investigates a current issue or event, offering different perspectives to local stories and revealing how it all affects you 6 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 11:55 – 19:36 (‘Health claims by MLM products’) & (‘Can Ganoderma cure or prevent cancer?’). 7 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 2:44 (‘Why MLMs have a bad reputation’). 8 Steven Alan Hassan, The BITE Model of Authoritarian Control: Undue Influence, Thought Reform, Brainwashing, Mind Control, Trafficking and the Law (PhD thesis, Fielding Graduate University 2020) 18– 20.https://www.researchgate.net/publication/348419785_THE_BITE_MODEL_OF_AUTHORITARIAN_CON TROL_UNDUE_INFLUENCE_THOUGHT_REFORM_BRAINWASHING_MIND_CONTROL_TRAFFICKI NG_AND_THE_LAW_A_dissertation_submitted accessed 15 June 2026.

318

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

resulted from months of torture, allegedly as a form of punishment at the instructions of her MLM coach for unsatisfactory sales performance. 9 While an isolated and extreme incident, it reflects broader concerns documented throughout the empirical and psychological literature regarding social coercion10 and the vulnerability of participants operating in an MLM.11 These concerns are significant given the size of the industry operating lawfully in Singapore, with a few hundred thousand distributors12 and approximately USD 591 million in sales in 2024 alone.13 The recent developments prompt the question: To what extent does Singapore’s regulatory framework effectively distinguish lawful MLM schemes from unlawful pyramid schemes, and does it provide meaningful protection to participants from financial and psychological harm? This large industry is governed by two statutes: the Multi-Level Marketing and Pyramid Selling (Prohibition) Act 197314 (“Prohibition Act”) and the Multi-Level Marketing and Pyramid Selling (Excluded Schemes and Arrangements) Order15 (“Exclusion Order”). The Prohibition Act broadly criminalises pyramid selling schemes, while the Exclusion Order permits specified commercial arrangements that satisfy prescribed statutory safeguards. In principle, the framework seeks to distinguish legitimate commercial arrangements from recruitment-driven schemes by focusing primarily on the formal source of participants’ benefits while balancing consumer protection against commercial freedom.

Selina Lum, ‘Man Admits to Torturing Woman, 19, to Death; Her Brother and Their MLM Coach Facing Murder Charge’ The Straits Times (9 February 2026) https://www.straitstimes.com/singapore/courtscrime/man-admits-torturing-girl-19-to-death-her-own-brother-and-their-mlm-coach-facing-murder-charge accessed 15 June 2026. 10 AWARE Singapore (@awaresingapore), ‘Was Herbalife seller Huang Baoying under coercive control before she died?’ (Instagram, 3 March 2026) https://www.instagram.com/p/DVaydoZAYIF 11 Steven Alan Hassan, The BITE Model of Authoritarian Control: Undue Influence, Thought Reform, Brainwashing, Mind Control, Trafficking and the Law (PhD thesis, Fielding Graduate University 2020) 18– 20.https://www.researchgate.net/publication/348419785_THE_BITE_MODEL_OF_AUTHORITARIAN_CON TROL_UNDUE_INFLUENCE_THOUGHT_REFORM_BRAINWASHING_MIND_CONTROL_TRAFFICKI NG_AND_THE_LAW_A_dissertation_submitted accessed 15 June 2026. 12 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 3:47-3:53 (‘Why MLMs have a bad reputation’). 13 World Federation of Direct Selling Associations, WFDSA STATS: 2024 Global Annual Direct Selling Statistical Data Report (November 2025) 47 < https://wfdsa.org/wp-content/uploads/2025/12/WFDSA-STATSReport-2024-2025-V1.pdf> accessed 15 June 2026 14 Multi-Level Marketing and Pyramid Selling (Prohibition) Act 1973 (2020 Rev Ed). 15 Multi-Level Marketing and Pyramid Selling (Excluded Schemes and Arrangements) Order (Cap 190, O 1, 2002 Rev Ed).

319

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

The difficulty, however, is that contemporary MLM structures may satisfy these formal safeguards while still exhibiting many of the economic and behavioural characteristics traditionally associated with pyramid schemes. A compensation structure may be presented as rewarding product sales rather than recruitment, yet remain economically dependent upon continual network expansion that resembles an unlawful pyramid scheme. 16 In other words, the statutory distinction may regulate the form through which compensation is received without addressing the underlying economic structure that generates participant harm. This article argues that the Singapore legal framework, while formally coherent, is substantively inadequate because it regulates the form of compensation rather than the structural reliance on recruitment, allowing pyramid-like schemes to operate lawfully. This article will first establish the conceptual distinction between pyramid schemes and MLMs before examining the legislative history of the Prohibition Act and the Exclusion Order. It then analyses the Singapore High Court’s decision in Chua Hock Soon James v Public Prosecutor17 to illustrate how the statutory safeguards operate in practice. The article argues that those safeguards inadequately capture product-based pyramid schemes because they regulate the formal source of compensation rather than whether the business remains economically dependent upon recruitment. It next compares Singapore’s approach with the U.S.’ s economic reality framework, which more effectively identifies recruitment-driven structures despite formal compliance. It concludes by proposing an economic reality assessment, mandatory earnings disclosures, licensing and competency requirements for recruiting distributors, and broader regulation of indirect recruitment incentives. II.

CONCEPTUAL FRAMEWORKS A. Pyramid Scheme

A pyramid scheme is a business model in which participants pay money to the scheme promoter in exchange for the right to receive compensation derived from the sale of goods or services (“commodities”), or from the recruitment of other persons into the scheme, or from the action(s) of their recruit(s). 18 However, compensation is derived primarily from recruitment

16 17 18

[2017] SGHC 230 Multi-Level Marketing and Pyramid Selling (Prohibition) Act 1973, s 2(1)(a)–(c).

320

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

rather than from the sale of commodities. 19 This business model is designed to enrich its founders quickly. It is effective at serving its purpose because of its deliberate design, which funnels money to the top by incentivising participants to recruit others, thereby driving rapid expansion. A pyramid scheme typically experiences exponential growth due to its inherent multiplication structure, which I will illustrate briefly: Person A pays $2000 to join a pyramid scheme and receives $100 for either selling a commodity or recruiting a new participant. Person A recruits four members; each of them recruits four new members, and so on. The number of participants grows exponentially – four recruits become sixteen, and sixteen become 64 in only three levels. Income for early participants correspondingly increases as the network expands, creating strong incentives for recruitment. However, from an economic perspective, pyramid schemes are unsustainable because revenue is generated primarily by buy-ins from new recruits rather than by genuine sales of goods. Consequently, the scheme inevitably collapses when it runs out of willing participants, resulting in widespread financial harm as most participants cannot recover their entry payments. An important distinction to make is that pyramid schemes are harmful not because of their recruitment, but because they are structurally dependent on it. B. MLMs MLM is a marketing program in which participants pay money to the program promoter in return for the right to (1) recruit additional participants as their downline; (2) resell commodities; and (3) receive compensation; provided that: (a) the payments received by each program participant are derived primarily from sales commodities, and not from recruitment (b) the marketing program has instituted and enforces rules to ensure that it is not a plan in which participants earn profits from recruitment rather than sales of commodities.20 While companies such as Amway and Herbalife are colloquially referred to as “MLM companies”, they are technically direct-selling companies that use an MLM compensation

Jon M Taylor, When Should an MLM or Network Marketing Program Be Considered an Illegal Pyramid Scheme? (Consumer Awareness Institute, 2000) 5, subheading ‘Pyramid Scheme’ http://www.pyramidschemealert.org/PSAMain/regulators/PPSdefined.pdf 20 Federal Trade Commission, Business Guidance Concerning Multi-Level Marketing (26 April 2024) subheading ‘What is multi-level marketing?’ https://www.ftc.gov/business-guidance/resources/businessguidance-concerning-multi-level-marketing#_ftnref1

321

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

structure.21 The term “direct selling” refers to the distribution of products directly to consumers outside conventional retail channels, 22 whereas “MLM” describes the underlying business model by which distributors may earn commissions from both their own sales and the sales of distributors within their downline. For convenience, this article uses the term “MLM companies” to refer to direct selling companies operating under an MLM business model. On preliminary observation, MLM companies appear structurally different from pyramid schemes. Unlike pyramid schemes, MLM companies typically offer genuine commodities for sale, compensate participants based on commodity sales, and frame recruitment as business expansion. In this respect, MLMs resemble legitimate franchise or distribution arrangements, which explains why some MLM companies may qualify for exclusion under Singapore’s Exclusion Order and operate lawfully. 23 Nevertheless, those familiar with how lawful MLM companies operate may opine that both schemes are deceptively similar, because they often are. Despite their formal differences, MLM schemes continue to attract criticism for exaggerated product claims, misleading earnings representations, and disproportionately poor outcomes for most participants.24 These concerns suggest that the distinction between pyramid schemes and MLMs may not be as clear-cut in practice as it appears in law. To answer this article’s research question, it is first necessary to examine the legal framework that distinguishes lawful MLM schemes from unlawful pyramid schemes, namely the Prohibition Act and the Exclusion Order. III.

SINGAPORE LEGISLATIVE FRAMEWORK A. Prohibition Act: broad prohibition

The Prohibition Act treats MLMs and Pyramid Schemes as synonymous, criminalising both. On preliminary observation, it is clear that Parliament deliberately cast the net very widely to

Direct Selling Association of Singapore, ‘FAQs – What is the difference between direct selling and multilevel marketing?’ https://dsas.org.sg/about-us/faqs/ accessed 15 June 2026. 22 Direct Selling Association of Singapore, ‘FAQs – What is direct selling?’ https://dsas.org.sg/about-us/faqs/ accessed 15 June 2026. 23 Multi-Level Marketing and Pyramid Selling (Excluded Schemes and Arrangements) Order (Cap 190, O 1, 2002 Rev Ed). 24 Jon M Taylor, The Case (for and) Against Multi-Level Marketing (submitted to the Federal Trade Commission, 2008) appendix 7A https://www.ftc.gov/sites/default/files/documents/public_comments/traderegulation-rule-disclosure-requirements-and-prohibitions-concerning-business-opportunities-ftc.r51199300008%C2%A0/00008-57281.pdf accessed 15 June 2026

322

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

capture as many potentially harmful schemes as possible, based on the broad words of Section 2(1) of the Prohibition Act, which reads: “pyramid selling scheme or arrangement” means any scheme or arrangement for the distribution or the purported distribution of a commodity by which — (a) a person may in any manner acquire a commodity or a right or a licence to acquire the commodity for sale, lease, licence or other distribution; (b) that person receives any benefit, directly or indirectly, as a result of — i.

the recruitment, acquisition, action or performance of one or more additional participants in the scheme or arrangement; or

ii.

the sale, lease, licence or other distribution of the commodity by one or more additional participants in the scheme or arrangement; and

(c) any benefit is or may be received by any other person who promotes, or participates in, the scheme or arrangement (other than a person referred to in paragraph (a) or an additional participant referred to in paragraph (b)).

Singapore’s zero-tolerance approach towards pyramid schemes is also reflected in the Explanatory Statement25 of the Prohibition Act. It essentially identified the two main elements of a pyramid scheme, as discussed in section 2.1, and explained that it ought to be banned in Singapore. The statutory framework, therefore, begins from a position of broad prohibition: MLMs and pyramid schemes are treated identically, and any scheme exhibiting the characteristics set out in Section 2(1) of the Prohibition Act is prima facie unlawful unless it falls within a subsequently created exclusion. B. Legislative intent behind the Prohibition Act The question that follows is why Parliament chose such an expansive and uncompromising approach. The legislative history demonstrates that Parliament regards pyramid selling not merely as an undesirable commercial practice, but as a serious social and economic threat that warrants outright prohibition. During the Second Reading of the Bill, 26 the then-Minister for

Multi-Level Marketing and Pyramid Selling (Prohibition) Bill (Bill No 45/1973, Singapore), Explanatory Statement, para 1. 26 Singapore Parliamentary Debates, Official Report (28 August 1973) vol 32, (Mr Hon Sui Sen, Minister for Finance) speaking on the Multi-Level Marketing and Pyramid Selling (Prohibition) Bill.

323

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Finance, Mr Hon Sui Sen, explained the concept of pyramid selling, or as it was “sometimes more euphemistically described, multi-level marketing”, and that it was already causing widespread financial ruin in jurisdictions such as the United States and the United Kingdom, with attempts at regulation having met with varying degrees of success. 27 Parliament was determined that Singapore should not suffer the same widespread financial harm and resolved to intervene before such schemes became entrenched domestically. In explaining the rationale behind the Bill, Mr Hon elaborated on the objectionable characteristics typically associated with pyramid schemes. First, such schemes relied on highpressure salesmanship that glossed over product marketability while conversely stressing the mostly illusory high rewards available. 28 Secondly, they misrepresented the economic realities faced by participants. Most participants were ultimately left with the burden of holding and disposing of unsold stock after making significant initial investments when they proved unable to sell the products.29 As a consequence, early entrants and promoters stood to benefit disproportionately at the expense of later participants, rendering the system inherently exploitative. It was concluded that such schemes, tainted by dishonesty and potentially fraudulent in intent, were objectionable and undesirable to the Government. 30 The legislative response in Singapore was therefore deliberately prophylactic: the objective was to eliminate rather than regulate such schemes. 31 Nevertheless, Parliament recognised the practical difficulty of distinguishing outright pyramid schemes from legitimate commercial arrangements exhibiting some pyramid element. 32 Direct selling companies whose primary source of income was the sale of commodities, rather than recruitment rewards, were “not necessarily objectionable and which should be allowed to develop.”33 Accordingly, considerable emphasis was placed on crafting a statutory definition that was both readily comprehensible and capable of effective enforcement. This balancing exercise took the form of Section 11 of the original Prohibition Act, 34 which preserved

Singapore Parliamentary Debates, Official Report (28 August 1973) vol 32, (Mr Hon Sui Sen, Minister for Finance) speaking on the Multi-Level Marketing and Pyramid Selling (Prohibition) Bill col 1284. 28 Multi-Level Marketing and Pyramid Selling (Prohibition) Bill (n 25) col 1286. 29 Ibid. 30 Multi-Level Marketing and Pyramid Selling (Prohibition) Bill (n 25) col 1287. 31 Ibid. 32 Multi-Level Marketing and Pyramid Selling (Prohibition) Bill (n 25) col 1288. 33 Multi-Level Marketing and Pyramid Selling (Prohibition) Bill (n 25) col 1288. 34 Multi-Level Marketing and Pyramid Selling (Prohibition) Act (Act 50 of 1973, Singapore) s 11.

324

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

flexibility by empowering the Minister to prescribe, by regulations, the conditions under which schemes or arrangements may lawfully be promoted in Singapore. This ministerial power served as a limited safeguard against the overly broad legislative provision. C. 2000 Prohibition Act Amendment and Exclusion Order Nearly three decades later, the broad prohibitionist framework adopted in 1973 eventually proved inadequate. By 2000, Parliament considered that the existing statutory definition of pyramid selling was “too narrow” due to its rigidities, such as requiring a participant to share his commission with another participant before the scheme was deemed pyramid selling. 35 Furthermore, it was insufficiently responsive to technological and business-environment developments.36 The amendment therefore sought to introduce a broader and more general definition capable of capturing objectionable schemes that would otherwise escape the Prohibition Act. Parliament nevertheless remained mindful that a broader definition increased the risk of unintentionally capturing legitimate commercial arrangements such as the insurance industry, which at the time was undergoing regulated expansion under a separate regulatory framework as Singapore sought to develop into a regional insurance hub. 37 While the Prohibition Act was broadened, Parliament simultaneously introduced the first Exclusion Order, exempting insurance businesses approved by MAS, qualifying master franchise schemes, and MLM companies that satisfy the following key safeguards: 38 (1) The benefit(s) received by participants result from the sale of a product, not from recruiting additional participants. (2) Promoters must not intentionally make misrepresentations relating to the scheme or product(s).

Singapore Parliamentary Debates, Official Report (9 May 2000) vol 72, (Mr Lim Swee Say, Minister of State for Trade and Industry) col 177. 36 Ibid col 177. 37 Ravi Menon, ‘Singapore as a Global Insurance Marketplace’ (Keynote Address, 12th Singapore International Reinsurance Conference, 6 November 2013) Monetary Authority of Singapore https://www.mas.gov.sg/news/speeches/2013/singapore-as-a-global-insurance-marketplace accessed 15 June 2026. 38 Multi-Level Marketing and Pyramid Selling (Excluded Schemes and Arrangements) Order (2002 Rev Ed) ss. 2(1)(c)(ii)-(vii).

325

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

(3) Promoters must not make any representation on the benefits of the scheme other than the sale of a product. (4) There should be proper records relating to earnings and benefits (5) There should be a clearly stated policy on refunds or buyback guarantees. The reforms also represented a shift in Singapore’s regulatory approach. Rather than prohibiting pyramid schemes/MLMs completely, Parliament implemented safeguards, recognising that certain multi-level distribution structures, such as the insurance industry, could serve legitimate commercial and economic objectives and ought not to be prohibited solely because of their pyramidal characteristics. However, the shift also creates a boundary zone between lawful MLM schemes and unlawful pyramid schemes. MLM arrangements may satisfy the formal requirements of the Exclusion Order while retaining economic dynamics commonly associated with pyramid schemes. It is this boundary zone, occupied by many established MLM companies, that this article examines. The current leading Singapore authority on pyramid selling, Chua Hock Soon James v PP (“Chua v PP”), provides the principal judicial interpretation of the Prohibition Act and Exclusion Order, and therefore forms the doctrinal starting point for that inquiry. IV.

CHUA HOCK SOON JAMES V PP A. Brief facts

The brief facts relate to the Appellants’ appeal against the orders of conviction. The Appellants were convicted of promoting a pyramid scheme. They contended that their scheme, the Global Edupreneur Program (“GEP”), was a franchising arrangement that qualified as an excluded scheme under Section 2(1)(b) of the Exclusion Order. The central issue in the appeal was whether the scheme fell within the Exclusion Order. The Court dismissed the Appellants’ appeal and held that they operated a pyramid scheme prohibited under the Prohibition Act and that it did not meet the requirements to be an excluded scheme under the Exclusion Order. Importantly, this was not a sophisticated MLM scheme that falls within the boundary zone. Rather, it was a relatively straightforward pyramid scheme which failed to satisfy the threshold requirements of the Exclusion Order.

326

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

B. Interpretation of the Prohibition Act The High Court interpreted the Prohibition Act consistently with Parliament’s legislative intent. Chan J held that a pyramid scheme, determined by a fact-driven examination, requires at least three tiers of participants: the original participant, an additional participant recruited by him, and a further participant recruited by the additional participant. 39 It criminalises schemes in which participants receive benefits, directly or indirectly, as a result of either recruiting another participant into the scheme or selling the commodity by one or more participants. As GEP’s participants paid licence fees, recruited additional participants, and received overriding commissions from their downlines' activities, 40 it constituted a pyramid scheme under the Prohibition Act. C. Interpretation of Exclusion Order The Court next considered whether it qualified as an excluded scheme under Section 2 of the Exclusion Order. The principal issue was whether Section 2(1)(c)(iii) should be interpreted independently or together with Section 2(1)(c)(ii), which prohibits recruitment-based benefits.41 Chan J considered three possible interpretations before adopting the “modified second interpretation”. 42 Under this approach, Sections 2(1)(c)(ii) and (iii) are read conjunctively. Accordingly, overriding commissions are permissible only when connected to the sale of commodities. Benefits arising merely from the recruitment of additional participants remain prohibited. The Court considered this interpretation most consistent with Parliament’s objective43 of distinguishing legitimate commercial activity from recruitment-driven schemes. Applying this interpretation, the Court held that the GEP contravened Sections 2(1)(c)(iii) 44 and (iv).45 Overriding commissions were calculated by reference to the recruitment of additional participants rather than the sale of educational programmes. The court rejected the Appellants’ argument that the relevant commodity was the licence to market the scheme’s programmes, holding that the licensing arrangements were merely incidental to a recruitment-

Chua v PP at [49] Chua v PP at [55] – [62] 41 Chua v PP at [98] – [100] 42 Chua v PP at [105], [113], [127]. 43 Chua v PP at [105],[127] 44 Chua v PP at [129] – [130] 45 Chua v PP at [137] – [139] 40

327

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

driven scheme. The duration of each licence depended upon recruiting new participants, while the licensing agreement and preview seminars placed primary emphasis on recruitment rather than product sales. Accordingly, the overriding commissions arose directly from recruitment, placing the scheme outside the Exclusion Order. D. Limits of Chua The significance of Chua v PP should not be overstated. The court was not confronted with a sophisticated MLM scheme that formally complied with the Exclusion Order while remaining economically dependent on recruitment. Instead, the GEP failed the Exclusion Order at a relatively early stage. The Appellants argued that the GEP was a franchise scheme falling within Section 2(1)(b) of the Exclusion Order, which requires compliance only with sections (ii), (iii), (iv) and (vi). By contrast, other schemes, including MLMs falling within Section 2(1)(c), must satisfy the full catalogue of conditions in sections (i) to (viii). The court expressly declined to determine whether the GEP was a franchise scheme. Chan J considered this issue unnecessary because, irrespective of its classification, the GEP already failed Sections 2(1)(c)(iii) and (iv). The convictions would therefore stand either way. 46 Consequently, the leading Singapore authority on pyramid selling does not examine the more difficult question of whether an MLM scheme that formally satisfies the Exclusion Order may nevertheless remain economically recruitment-driven. Nor does it consider the operation of safeguards such as the requirements relating to audited earnings disclosures, misleading representations or coercive conduct. Those questions are therefore left to be decided in future cases and, more fundamentally, by Parliament. Although the court ultimately declined to determine whether the GEP constituted a franchise, its discussion of section 2(1)(c)(iii) reveals an important feature of Singapore’s legislative design. Parliament did not regard recruitment itself as inherently objectionable. Rather, the critical distinction lies in whether the economic benefit arises from genuine commercial activity or from expanding the participant network. Immediate recruitment commissions may therefore be permissible where they are merely incidental to a legitimate commercial

Chua v PP at [152].

328

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

arrangement, such as franchising, whereas rewards economically dependent upon continual network expansion remain prohibited. The modified second interpretation therefore reinforces Parliament’s broader objective of distinguishing legitimate commerce from recruitment-driven compensation rather than prohibiting hierarchical structures as such. E. Boundary zone The law uses formal characteristics as proxies for economic substance. This works reasonably well for traditional franchise arrangements, insurance businesses and financial advisory services, where recruitment ordinarily facilitates genuine commercial activity. However, MLMs occupy a more uncertain space. MLMs may formally derive benefits from product sales while remaining economically dependent on the continual recruitment of new participants and the purchases made by those participants. In such circumstances, the distinction between sales and recruitment becomes increasingly blurred. Recruitment may no longer facilitate the sale of commodities; rather, commodity sales may become the mechanism through which recruitment is monetised. MLMs may therefore exist within the boundary created by the Exclusion Order. They are not necessarily unlawful pyramid schemes of the kind considered in Chua v PP, nor are they legitimate franchise arrangements whose commercial value is self-evident. The central question is therefore not whether recruitment is present, but whether the scheme remains economically viable in the absence of continual network expansion. It is this distinction between formal compliance and economic reality that the next section examines. V.

FORM VS SUBSTANCE: MLMS WITHIN THE BOUNDARY A. Core doctrinal problem

Singapore distinguishes lawful MLMs from unlawful MLMs/pyramid schemes by asking whether participants receive benefits from product sales rather than recruitment. However, as mentioned above, this method is ineffective because of the blurred distinction. The issue is not the formal source of participants’ benefits, but whether the scheme is economically sustained by genuine retail demand or by continual recruitment. A compensation plan may satisfy the statutory requirement that rewards arise from product sales while remaining structurally dependent upon recruitment-driven purchases. 329

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

B. MLMs on paper On paper, MLM companies resemble legitimate retail distribution businesses. Participants earn commissions from product sales, expand their businesses by recruiting additional distributors, and are commonly described as independent business owners rather than employees. Unlike conventional employment, distributors are not paid salaries but operate as independent contractors whose income depends upon sales performance. MLM companies compare themselves to franchise businesses. Both models involve training, business expansion and the distribution of products through independent operators. Recruitment is therefore presented as a commercially legitimate means of expanding product distribution rather than as an independent source of remuneration. The MLM structure appears consistent with the distinction recognised in the Exclusion Order. Participants receive commissions tied to product sales, and recruitment is typically characterised as a means of expanding product distribution rather than an independent source of remuneration. This explains why MLM companies, like franchising business models, may qualify for exclusion and be permitted to operate despite their pyramidal structure. C. MLMs in practice However, the formal presentation of MLM companies may not reflect their operational reality. Commercial success seems more dependent on continually expanding one’s downline and encouraging further participation in the compensation structure than on developing a sustainable retail customer base. First, product sales are often driven by internal consumption rather than by independent retail demand. Instead of genuine consumer sales, a significant proportion of products are purchased by participants to satisfy their monthly purchase thresholds, maintain eligibility for commissions, qualify for rank advancement or remain active within the compensation plan. 47 While these transactions are recorded as product sales, they are generated by participation in the scheme rather than by independent consumer demand.

Claudia Groß and Dirk Vriens, ‘Buy! Buy! Buy!—How Multilevel Marketing Companies Pressure Their Participants to Buy Their Products’ (2025) 44(4) Journal of Public Policy & Marketing 559 https://journals.sagepub.com/doi/epub/10.1177/07439156241301737 accessed 15 June 2026.

330

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Secondly, most uplines place greater emphasis on recruitment rather than selling the product. Investigative journalism has repeatedly documented recruitment meetings and promotional seminars in which prospective participants are encouraged to focus primarily on building a downline rather than developing a sustainable retail customer base. 48 Recruitment is commonly marketed as a pathway to financial independence, passive income and exponential earnings, with overriding commissions from one’s team often receiving greater emphasis than the commercial viability of selling products to end consumers. Thirdly, recruitment and product sales become economically intertwined. Each new distributor represents not merely an additional salesperson, but also a new purchaser of products and a potential source of further recruitment. Downline expansion therefore generates additional sales volume even where little additional retail demand exists. Consequently, product sales and recruitment no longer operate as distinct commercial activities; rather, recruitment itself becomes the mechanism through which product sales are generated. While these observations are not necessarily true for every MLM company, they demonstrate how a compensation structure may remain formally based on product sales while becoming economically dependent on continual recruitment. The critical issue is therefore not whether product sales occur, but how those sales are generated. D. Four methods of generating sales volume Previously, pyramid schemes generated revenue primarily through recruitment fees. Contemporary MLM schemes instead generate sales volume through product purchases. As illustrated in Figure 1, these purchases may arise through four distinct methods.

ABC News, ‘Herbalife Investigation: American Dream for Sale? and The Whistleblower’ (YouTube, 23 April 2014) 6:23 https://www.youtube.com/watch?v=EqOMHOgt3JQ accessed 15 June 2026.

331

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Figure 1

On preliminary review, Methods 1 and 2 appear normative and align with how MLM companies present themselves: as opportunities to build a genuine business without conventional business overheads. However, most distributors generate sales primarily through methods 3 and 4, resulting in financial losses, as commodities usually end up hoarded in their homes, commonly termed in the industry as “garage qualified” 49. When distributors exhaust their leads and money to purchase the commodities, they focus on recruiting new distributors as they benefit from their downline sales. Revenue from upfront recruitment fees that translate into recruitment benefits is replaced by revenue from the self-purchase of commodities that translate into benefits from sales performance. Self-purchases by distributors and new downlines are masked as genuine commercial activity, thereby complying with the Exclusion Order, which permits compensation only from sales. Once again, despite removing the benefits from recruitment, there is still an almost pure reliance on recruitment, similar to a pyramid

Commodities are often stored in a distributor’s garage for the supposed resale, although most distributors are usually unable to sell, leading to financial losses.

332

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

scheme. It is clear that MLM companies can circumvent the Exclusion Order by, ironically, satisfying it rather than breaching it. VI.

PPS FRAMEWORK

This phenomenon closely resembles what Taylor describes as a product-based pyramid scheme (“PPS”).50 A PPS is a pyramid scheme that offers a product, similar to a lawful MLM, except that rather than generating revenue primarily through upfront recruitment fees, participants purchase commodities—often for their own consumption or purported resale—thereby both financing the scheme and qualifying for commissions within the distributor hierarchy. Product purchases therefore replace recruitment fees as the scheme's economic engine while preserving its underlying recruitment-driven structure.51 Taylor argues that durable PPSs share several structural characteristics commonly exhibited by MLM companies. These characteristics provide a useful analytical framework for assessing whether Singapore’s Exclusion Order distinguishes genuine commercial activity from recruitment-driven enterprises. A. PPS Characteristic 1: A chaining hierarchy of distributors without functional or territorial limits The first characteristic is the presence of a multi-level distributor hierarchy that extends beyond what is functionally necessary for legitimate product distribution. Even the largest conventional distribution system in America can be covered by a maximum of five intermediary levels, which ordinarily exist because each performs an identifiable commercial function, such as wholesale distribution, regional management or retail sales.52 According to Taylor, anything more than that is bloated, making sales at a competitive retail markup unprofitable and unrealistic.53 In contrast, MLM structures permit seemingly unlimited succession of distributor levels without corresponding commercial justification. Participants are encouraged not merely to sell products but to recruit further distributors, creating an ever-expanding compensation chain. However, the excessive levels seldom have any functional justification beyond

Jon M Taylor, When Should an MLM or Network Marketing Program Be Considered an Illegal Pyramid Scheme? (Consumer Awareness Institute 2000) 18 https://www.pyramidschemealert.org/PSAMain/regulators/PPSdefined.pdf accessed 15 June 2026. 51 Ibid 6 52 Ibid 12 53 ibid

333

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

encouraging participants to recruit, 54 and the mathematical reality is insidiously glossed over while simultaneously emphasising mostly illusory large potential incomes—an element prominent in pyramid schemes.55 Additionally, uplines of many levels may be profiting to an extreme degree from the losses of those beneath them. 56 Such exorbitant incomes are the cumulative result of the commercial activities of hundreds or even thousands of downstream participants that MLM promoters refer to as “residual income” or “leverage”. Furthermore, unlike franchise systems, which commonly allocate operating territories, 57 MLM participants are ordinarily recruited without meaningful territorial or market constraints in a given area. This creates market saturation and the inevitable collapse of the scheme. These features operate together to create the illusion of unlimited earning potential while structurally concentrating gains at the top and distributing losses across most participants. 58 B. PPS Characteristic 2: Relative vertical equality masking extreme horizontal inequality Taylor’s second characteristic concerns what he terms “relative vertical equality” (“RVE”), whereby participants at different levels of the hierarchy appear to receive similar commission rates for comparable sales activity. This formal equality, however, masks what Taylor describes as “extreme horizontal inequality” (“EHI”), because participants occupying higher positions benefit from exponentially larger downline networks. Consequently, a small minority of distributors capture a disproportionate share of total rewards while the overwhelming majority earn little or incur losses. The apparent equality of opportunity masks substantial inequality of outcome. Although compensation is formally linked to product sales, meaningful income increasingly depends upon occupying advantageous positions within the recruitment hierarchy.

ibid Multi-Level Marketing and Pyramid Selling (Prohibition) Bill (n 25) col 1286. 56 Jon M Taylor, When Should an MLM or Network Marketing Program Be Considered an Illegal Pyramid Scheme? (Consumer Awareness Institute 2000) 18 https://www.pyramidschemealert.org/PSAMain/regulators/PPSdefined.pdf accessed 15 June 2026. 57 Franchising & Licensing Association (Singapore), ‘Singapore’s Franchise Laws Explained — What Every Franchisee and Franchisor Must Know’ (20 March 2026) (‘Essential Clauses’, paras 1 and 6) https://flasingapore.org/article/singapore-s-franchise-laws-explained-what-every-franchisee-and-franchisormust-know ⁠ accessed 15 June 2026. 58 Jon M Taylor, When Should an MLM or Network Marketing Program Be Considered an Illegal Pyramid Scheme? (Consumer Awareness Institute 2000) 18 https://www.pyramidschemealert.org/PSAMain/regulators/PPSdefined.pdf accessed 15 June 2026. 55

334

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

C. PPS Characteristic 3: Purchase and recruitment-based qualification requirements (“pay-to-play”) The third characteristic identified by Taylor concerns what he describes as the “pay-to-play” feature of product-based pyramid schemes.59 PPS masquerading as MLMs are often allowed to grow and flourish unchecked because they do not require a large upfront enrolment fee. Recruiters often do not profit from the initial recruitment of a distributor, thereby fulfilling Section 2(c)(iii) of the Exclusion Order. Instead, in many MLM programmes, distributors are required or strongly incentivised to purchase specified quantities of products, meet sales quotas, recruit additional distributors, or satisfy other qualification requirements before they become eligible for higher commission rates, bonus payments, or rank advancement. This creates a continual treadmill effect that keeps distributors recruiting or moving more products, typically by self-purchase. Although these requirements are formally linked to product sales, they frequently operate as participation costs necessary to remain competitive within the compensation structure. Taylor’s observations are reflected in the CNA Insider’s investigation. A former MLM participant (Kelly) explained that distributors were required to maintain a minimum monthly purchase of approximately 100 points (around $200) to remain eligible to earn commissions. After four years in the scheme, she had accumulated approximately $3,000 worth of essential oils because she had been “buying them faster than [she] was using them”. She ultimately concluded that she had spent more on products than she had earned. 60 Kelly’s continued purchases were driven by the need to remain eligible under the compensation structure. The scheme therefore created an ongoing incentive to purchase products irrespective of genuine demand, illustrating the “continual treadmill” described by Taylor.61 Taken together, these characteristics demonstrate that, notwithstanding formal compliance with statutory requirements, the economic reality of PPS remains structurally dependent on recruitment-driven expansion. In this way, such MLMs do not circumvent the safeguards by 59

Ibid 16. CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 10:40 (‘How much MLM distributors really earn’). 61 Jon M Taylor, When Should an MLM or Network Marketing Program Be Considered an Illegal Pyramid Scheme? (Consumer Awareness Institute 2000) 16 https://www.pyramidschemealert.org/PSAMain/regulators/PPSdefined.pdf accessed 15 June 2026. 60

335

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

breaching their requirements but by satisfying them. Moreover, Singaporeans’ high trust and positive attitude towards the legal system 62 may be exploited by promoters who invoke the legality of permitted PPS, claiming “since it is legal, it is alright,” to lure participants. VII.

EFFECTIVENESS OF SINGAPORE SAFEGUARDS

The effectiveness of Singapore’s statutory safeguards should be assessed against the legislative objectives they were intended to achieve. This section evaluates the safeguards against three interrelated indicators derived from Parliament’s legislative intent and the PPS framework developed above: whether they prevented widespread financial harm, eliminated structural dependence on recruitment, and removed the objectionable features historically associated with pyramid selling. A. Financial harm Data consistently show that the overwhelming majority of MLM participants incur net financial losses after accounting for product purchases and operating expenses. Approximately 99% of participants ultimately lose money, while income disclosure statements published by MLM companies such as Amway and Herbalife demonstrate that earnings remain heavily concentrated amongst a small minority of participants, with most distributors receiving little or no net income after expenses. 63 Street interviews conducted by CNA similarly reflect negative experiences of MLMs in Singapore, with people associating them with scams. 64 These outcomes closely resemble the structural characteristics identified by Taylor’s PPS framework, suggesting that compliance with product-based compensation requirements has not fundamentally changed the economic reality of pyramid selling. Instead of paying upfront recruitment fees, participants increasingly incur losses through qualification purchases and internal consumption. Product purchases essentially perform the same economic function as recruitment fees did previously. It generates revenue from participant expenditure rather than 62

Speech by Minister for Law, K Shanmugam, during the Committee of Supply Debate 2016 (Ministry of Law, 6 April 2016) para 3(a)–(d) https://www.mlaw.gov.sg/news/parliamentary-speeches/speech-by-minister-for-law-k-shanmugam--during-the-committee-of/ accessed 15 June 2026. 63 Jon M Taylor, The Case (for and) Against Multi-Level Marketing (submitted to the Federal Trade Commission, 2008) appendix 7A https://www.ftc.gov/sites/default/files/documents/public_comments/traderegulation-rule-disclosure-requirements-and-prohibitions-concerning-business-opportunities-ftc.r51199300008%C2%A0/00008-57281.pdf accessed 15 June 2026 64 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 2:44 (‘Why MLMs have a bad reputation’).

336

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

genuine consumer demand. While the legal form of the transaction has therefore changed, the economic consequence remains substantially the same for most participants. B. Structural dependence on recruitment While the Exclusion Order first appears to have eliminated the structural dependence on recruitment that characterises pyramid selling, the distinction may not fully capture the economic reality of contemporary MLM schemes. Recruitment no longer generates a one-off participation fee. Instead, it produces continuing product purchases through the selfconsumption and qualification purchases of newly recruited distributors. Consequently, recruitment may continue to be the principal economic driver of the scheme notwithstanding formal compliance with Section 2(1)(c)(iii) Exclusion Order. The result is that a scheme may comply with the statutory prohibition on recruitment-based rewards while remaining structurally dependent upon continual network expansion to sustain sales volume. C. Objectionable elements First, the statutory prohibition against misleading representations does not appear to have eliminated unrealistic earnings expectations. Investigative reporting has repeatedly documented exaggerated income claims, including representations that participants could earn up to US$42,000 per month, 65 despite Herbalife’s own income disclosure statement showing that only about 0.3% of distributors earned more than US$250,000 annually. 66 Likewise, CNA Insider reported that average monthly earnings for direct-selling participants in the United Kingdom and the United States were approximately US$827.10 and US$657, respectively, before business expenses.67 Large motivational conventions and selective success stories continue to reinforce the impression that such outcomes are broadly attainable, 68 even though the empirical evidence suggests otherwise. Secondly, exaggerated product representations likewise remain prevalent despite the statutory prohibition against misleading claims. Herbalife distributors were documented claiming that 65

ABC News, ‘Herbalife Investigation: American Dream for Sale? and The Whistleblower’ (YouTube, 23 April 2014) 7:03 https://www.youtube.com/watch?v=EqOMHOgt3JQ accessed 15 June 2026. 66 Ibid 7:16 67 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 7:14 (‘How much MLM distributors really earn’). 68 Ibid 8:15.

337

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

their products could assist with serious medical conditions, 69 while CNA Insider similarly identified distributors promoting products using unsubstantiated health claims. 70 The persistence of such representations suggests that the statutory safeguards have not fully prevented the misleading promotional practices Parliament sought to eliminate. Finally, the evidence indicates that high-pressure recruitment and social coercion remain in many MLM environments. Former participants have described sustained pressure to prioritise recruitment, attend motivational conventions, devote significant time to the business, and distance themselves from individuals who express scepticism. 71 A particular concern is that some MLM companies are known to target vulnerable individuals, including young mothers, pregnant women and persons experiencing financial hardship, with promises of financial freedom, flexible working arrangements and a supportive community.72 Although Huang Baoying’s case was exceptional and attributable to criminal conduct rather than the MLM model itself, AWARE’s observations regarding financial control, social isolation and coercive influence73 reflect forms of pressure documented more broadly within the MLM literature. The case therefore illustrates the potential severity of these dynamics without suggesting that they are inherent in every MLM arrangement. Taken together, the evidence suggests that the statutory safeguards have not eliminated many of the objectionable features Parliament associated with pyramid selling. Rather than disappearing, those practices appear to have adapted to operate within the formal boundaries of the Exclusion Order. The safeguards therefore regulate the legal form of compensation more effectively compared to the commercial practices wherein recruitment is sustained, and sales volume is generated. D. Counterargument

ABC News, ‘Herbalife Investigation: American Dream for Sale? and The Whistleblower’ (YouTube, 23 April 2014) 9:25 https://www.youtube.com/watch?v=EqOMHOgt3JQ accessed 15 June 2026. 70 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 11:55 – 19:36 (‘Health claims by MLM products’) & (‘Can Ganoderma cure or prevent cancer?’). 71 BBC deepwatch, ‘The Secrets Of Making Money On Social Media Through Multi-Level Marketing: Ellie Undercover’ (YouTube, 5 May 2019) https://www.youtube.com/watch?v=r4YGQ0l5UeE 6:15 accessed 15 June 2026. 72 BBC Stories, ‘MLM: Are Make-up Empires Exploiting Mums?’ (YouTube, 26 April 2019) https://www.youtube.com/watch?v=o5xhNXVfPYQ accessed 15 June 2026. 73 AWARE Singapore (@awaresingapore), ‘Was Herbalife seller Huang Baoying under coercive control before she died?’ (Instagram, 3 March 2026) https://www.instagram.com/p/DVaydoZAYIF

338

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Nevertheless, one possible argument is that the existing safeguards have been effective. In response to Parliamentary Questions concerning MLM regulation, 74 the Government observed that CASE75 had received, on average, fewer than ten complaints annually relating to MLM schemes over the preceding five years, despite receiving approximately 24,000 consumer complaints each year. The Minister therefore stated that there were no plans to review the existing legislative framework. 76 On one view, the comparatively low complaint rate may suggest that participant harm is uncommon and that the safeguards contained in the Exclusion Order are functioning as intended. However, complaint statistics are not necessarily an accurate proxy for participant harm. During CNA Insider’s investigation, the Vice Chairman of the Direct Selling Association of Singapore acknowledged that complaints relating to MLM schemes were relatively uncommon, but suggested that participants were often “too embarrassed to complain”.77 He further observed that sophisticated operators were familiar with Singapore’s legal framework and understood how to structure their businesses so as to remain compliant with the law. 78 Low complaints may reflect barriers to reporting and legal compliance rather than the absence of harmful practices. Singapore’s high level of public confidence in the legal system 79 may itself suppress complaints. Where promoters emphasise that a scheme satisfies the Exclusion Order, participants may reasonably infer that Parliament has already determined the model to be sufficiently safe. Participants who subsequently incur losses may attribute their experience to personal failure rather than structural features of the compensation system, while others may feel embarrassed to complain after recruiting family members or friends into the scheme.

Singapore Parliament, Written Answers to Questions, ‘Outcome of Illegal Multi-level Marketing Scheme Cases’ (3 March 2023) (Mr K Shanmugam, Minister for Home Affairs). 75 The Consumers Association of Singapore (CASE) is a non-governmental organisation that is committed to protect consumer interests through information, education and the promotion of fair and ethical trade practices. 76 Singapore Parliament, Written Answers to Questions for Oral Answer Not Answered by End of Question Time, ‘Complaints against Multi-level Marketing Companies’ (4 July 2017) (Mr S Iswaran, Minister for Trade and Industry (Industry)). 77 CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 20:10 (‘Why companies get away with exaggerated claims’). 78 Ibid 20:17 (‘Why companies get away with exaggerated claims’). 79 Speech by Minister for Law, K Shanmugam, during the Committee of Supply Debate 2016 (Ministry of Law, 6 April 2016) para 3(a)–(d) https://www.mlaw.gov.sg/news/parliamentary-speeches/speech-by-minister-for-law-k-shanmugam--during-the-committee-of/ accessed 15 June 2026.

339

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

Therefore, complaint statistics risk understating participant harm and overstating the effectiveness of the existing safeguards. VIII.

LIMITS OF FORMAL SAFEGUARDS

The Exclusion Order contains safeguards extending beyond the prohibition on recruitment rewards. However, several provisions remain framed in a manner that emphasises formal compliance rather than substantive protection for participants. A. Fair and accurate records Section 2(1)(c)(v) of the Exclusion Order requires promoters to maintain fair and accurate records relating to participant earnings. While this promotes transparency, the safeguard regulates only the maintenance of records, not their disclosure. Nothing compels promoters to furnish prospective participants with those records before deciding whether to join the scheme, nor does the Exclusion Order prescribe any standardised earnings disclosure. A promoter may therefore fully comply with the safeguard while prospective participants continue to rely on selective testimonials, exceptional success stories, and optimistic earnings projections. B. Section 2(1)(c)(vi): Misleading representations Section 2(1)(c)(vi) similarly reflects Parliament’s concern with misleading promotional practices by prohibiting false or misleading representations concerning the scheme or its products. Yet the provision is primarily directed towards express misrepresentations and offers limited guidance on the subtler methods by which modern MLM schemes shape participants' expectations. Contemporary MLM recruitment does not rely on objectively false statements, but rather on technically possible yet incomplete representations that create unrealistic expectations regarding likely earnings and commercial success. Prospective participants are still exposed to luxury lifestyle imagery, high-profile conventions, and carefully curated social media content that collectively create unrealistic expectations regarding likely earnings or product efficacy. The statutory prohibition therefore risks addressing only the most obvious forms of deception while leaving more sophisticated forms of behavioural influence largely untouched.

340

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

The practical operation of section 2(1)(c)(vi) likewise presents enforcement difficulties. As discussed in Section VII, there continues to be documentation of widespread unsubstantiated health claims notwithstanding formal corporate policies prohibiting such conduct. The persistence of these representations suggests that prohibiting misleading statements as a matter of law is considerably easier than enforcing that prohibition across decentralised distributor networks. This concern is particularly significant given that wellness products accounted for approximately 62% of Singapore’s direct-selling retail sales in 2024. 80 Where purchasing decisions frequently depend upon representations concerning health benefits, ineffective enforcement may substantially undermine the protective purpose of Section 2(1)(c)(vi). While it may be effective in prohibiting misleading conduct as a matter of law, it is seemingly more difficult to operationalise in practice. These observations do not suggest that the safeguards are ineffective. Rather, they demonstrate that the Exclusion Order primarily regulates formal compliance rather than substantive protection for participants. A scheme may maintain accurate records without informing prospective participants, and avoid objectively false representations while continuing to cultivate unrealistic expectations through implication, selective disclosure and behavioural persuasion. The statutory inquiry therefore remains primarily directed to whether prescribed safeguards formally exist, rather than to whether they successfully distinguish legitimate direct selling from recruitment-driven commercial structures. The Exclusion Order therefore regulates legal form more effectively than economic substance. This distinction becomes particularly apparent when compared with the U.S, whose regulatory approach increasingly emphasises economic substance over formal compliance. IX.

UNITED STATES APPROACH A. FTC Framework

Unlike Singapore, the U.S does not distinguish lawful MLMs from unlawful pyramid schemes through detailed statutory exclusions. Instead, legality has developed through § 5 of the Federal Trade Commission Act (“FTC Act”), Federal Trade Commission (“FTC”) enforcement and judicial decisions. The starting point remains In re Koscot Interplanetary Inc. 81, in which the 80

World Federation of Direct Selling Associations, WFDSA STATS: 2024 Global Annual Direct Selling Statistical Data Report (November 2025) 45 < https://wfdsa.org/wp-content/uploads/2025/12/WFDSA-STATSReport-2024-2025-V1.pdf> accessed 15 June 2026 81 86 FTC 1106 (1975).

341

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

FTC defined a pyramid scheme as one in which participants pay consideration for the right to sell products and receive rewards for recruiting additional participants that are unrelated to sales to ultimate users. 82 This definition remains the foundation of the FTC’s approach. Rather than assessing an MLM’s legality solely by reference to prescribed safeguards, the FTC undertakes a fact-specific inquiry into the compensation structure as a whole. The FTC examines how the business operates in practice by considering factors such as the company’s marketing representations, participant experiences, compensation plan and incentives. 83 They also consider other nuanced factors, such as why individuals sign up, who purchases the products, and why they do. 84 Therefore, the FTC does not simply accept that compensation derived from product sales is necessarily lawful. Instead, it asks whether the apparent retail activity reflects genuine consumer demand or merely disguises recruitment-driven compensation. This inquiry has been progressively refined through judicial decisions. Webster v Omnitrition International Inc85 established that an MLM “cannot save itself simply by pointing to the fact that it makes some retail sales.”86 In other words, the mere engagement in some retail sales does not protect itself from liability. Similarly, the FTC rejects the proposition that an MLM is lawful merely because commissions are triggered by product sales 87 or because the company sells genuine products or services. 88 The relevant inquiry is whether the compensation structure, viewed as a whole, principally incentivises recruitment or genuine retail demand. Consequently, courts examine both the written compensation plan and the practical incentives it creates rather than its formal description.

Koscot at [59]. Federal Trade Commission, Business Guidance Concerning Multi-Level Marketing (Federal Trade Commission, 2024) Question 2 https://www.ftc.gov/business-guidance/resources/business-guidanceconcerning-multi-level-marketing accessed 13 July 2026. 84 Ibid Question 4. 85 79 F 3d 776 (9th Cir 1996) 782 https://caselaw.findlaw.com/court/us-9th-circuit/1139924.html accessed 13 July 2026. 86 ibid 79 F 3d 776 (9th Cir 1996) https://caselaw.findlaw.com/court/us-9thcircuit/1139924.html#:~:text=On%20its%20face%2C%20Omnitrition's%20program%20appears%20to%20be% 20a%20pyramid%20scheme.%20%E2%80%82%20Omnitrition%20cannot%20save%20itself%20simply%20by %20pointing%20to%20the%20fact%20that%20it%20makes%20some%20retail%20sales accessed 15 July 2026. 87 FTC (n83) Question 7. 88 FTC (n83) Question 6. 83

342

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

FTC v BurnLounge Inc89 further clarified this approach. The Ninth Circuit reaffirmed that courts must examine how the MLM business operates in practice rather than merely how its compensation plan is described. 90 An MLM therefore operates as a pyramid scheme where the “focus [is] in promoting the program rather than selling the products.”91 On the facts, BurnLounge’s business focused on promoting the Mogul programme rather than retailing products, with recruitment and recruitment rewards forming an integral part of its compensation structure because recruiting participants increased eligibility for rewards and further recruitment generated greater rewards. 92 Likewise, in FTC v Noland93, the District Court found that the MLM drove product sales by pushing recruitment to sell large upfront product purchases, large monthly qualification purchases to “stay on the path to financial freedom”, and duplication strategies that encouraged participants to induce their downlines to engage in the same conduct. 94 These cases demonstrate that the existence of product sales does not conclude the inquiry where the compensation structure itself incentivises recruitment. Notwithstanding Nolan, the FTC adopts the same substance-based approach when assessing participant purchases and internal consumption. Purchases by participants are neither automatically treated as evidence of genuine retail demand nor automatically disregarded. 95 Instead, the FTC examines whether the compensation structure encourages purchases for reasons other than satisfying genuine personal or retail demand. These factors include inventory loading, duplication strategies that encourage recruits to make substantial initial purchases, “preferred” customer arrangements that precede sign-up, and evidence that participants purchase products primarily to qualify for compensation rather than because they genuinely wish to consume them. 96 An example of the FTC’s assessment is illustrated in BurnLounge. BurnLounge argued that participants purchased product packages because they genuinely wanted the accompanying music-related products. However, after a Court ordered BurnLounge to separate its product packages from the business opportunity, monthly sales of those packages fell by nearly 98 per cent. 97 The overwhelming majority of purchases had

753 F 3d (9th Cir 2014). BurnLounge 878, 883 91 BurnLounge 884 92 BurnLounge 884-887 93 CV-20-00047-PHX-DWL, 2023 WL 3372517 94 FTC v Nolan at [43] 95 FTC (n83) Question 8 96 FTC (n83) Questions 8-9. 97 BurnLounge 885 90

343

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

therefore been motivated by the opportunity to earn income rather than genuine consumer demand. Hence, product sales cannot be viewed in isolation and must be understood within the broader operation of the compensation structure. The U.S. approach therefore distinguishes lawful MLMs from unlawful pyramid schemes through a fundamentally different inquiry from Singapore’s Exclusion Order. However, it is not without limitations. MLM participants continue to suffer substantial financial losses, enforcement remains largely case-specific, and uncertainty persists regarding the precise boundary between lawful MLMs and unlawful pyramid schemes. Nevertheless, its methodology is better equipped to deal with PPS. Rather than asking whether remuneration is formally attributable to product sales, it asks whether those sales are generated by genuine retail demand or by recruitment-driven incentives embedded in the compensation structure. Product sales, participant purchases and organisational safeguards are therefore not determinative in themselves. Each forms part of a holistic factual inquiry into whether recruitment, rather than retailing, constitutes the business's true commercial focus. X.

COMPARATIVE ANALYSIS

While Singapore and the U.S. adopt different regulatory approaches, they both seek to distinguish legitimate direct-selling businesses from unlawful pyramid schemes and protect participants from harm. The question, therefore, is whether their respective legal frameworks can identify recruitment-driven schemes without unnecessarily restricting legitimate business practice. Singapore’s Exclusion Order adopts a rules-based approach by prescribing certain conditions that lawful MLMs must satisfy.98 This provides greater legal certainty and commercial predictability for legitimate businesses. Whereas the U.S. adopts a functional inquiry directed towards the commercial reality of the compensation structure. As demonstrated throughout Section IX, the FTC does not treat any single safeguard as determinative. Instead, product sales, participant purchases, refund policies, and compensation terminology are taken as evidence and considered holistically, rather than as legal conclusions. Ultimately, it is a matter of whether the compensation structure is economically sustained by genuine retail demand or by continual recruitment. Consequently, the U.S. framework is inherently less certain than Singapore’s, but considerably more responsive to sophisticated

Multi-Level Marketing and Pyramid Selling (Excluded Schemes and Arrangements) Order (2002 Rev Ed) ss. 2(1)(c)(ii)-(vii).

344

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

MLM models (or PPS) that satisfy formal legal requirements while remaining recruitmentdependent in substance. Neither approach is superior in all respects; it simply reflects a different balance between legal certainty and regulatory sensitivity. The more significant comparative difference emerges when the two frameworks are tested against PPSs. As discussed above in Section VI, PPS rarely resemble traditional chain-referral schemes. Instead, they operate through genuine products, formally compliant compensation plans and extensive internal consumption, while remaining economically dependent upon continual recruitment. Measured against these characteristics, the FTC’s methodology more effectively captures the underlying economic structure of PPS by looking beyond product-based compensation and examining the reasons for sales. Product purchases motivated by qualification requirements or ‘progression’ within the compensation plan are treated as evidence that apparent retail demand may in reality be recruitment-driven. Likewise, the existence of genuine products or some retail sales does not preclude liability where the business opportunity continues to revolve around recruitment.99 Singapore approaches the same issue differently. Once the statutory safeguards prescribed by the Exclusion Order are satisfied, the legislation generally does not inquire whether product purchases are themselves generated by genuine consumer demand or by incentives embedded within the compensation structure. 100 Consequently, qualification purchases, internal consumption and inventory loading continue to sustain participant remuneration notwithstanding formal compliance with the statutory framework. The Exclusion Order regulates the legal source of participant benefits, but does not necessarily examine the commercial ‘legitimacy’ or ‘authenticity’ behind those benefits. To conclude, Singapore’s rules-based framework may provide greater certainty for both regulators and legitimate businesses in dealing with conventional MLM schemes. However, where compensation structures are deliberately designed to satisfy formal statutory safeguards while preserving recruitment-driven incentives, the U.S. methodology appears better equipped to identify the underlying economic reality. In particular, its approach to examining participant behaviour, internal consumption, qualification purchases and the practical operation of compensation plans enables it to identify PPS masquerading as MLM schemes that may

Omnitrition (n86) CNA Insider, MLMs: Business Opportunity Or A Scam? | Talking Point | Full Episode (YouTube, 31 July 2023) https://youtu.be/9bSV4HbpKfw?si=d0tcZFzwUnsg54NI&t=164 accessed 10 June 2026, 10:40 (‘How much MLM distributors really earn’). 100

345

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

otherwise appear legally compliant. Nevertheless, this does not suggest that Singapore should abandon its statutory framework nor attempt to completely transplant the U.S. model. The certainty provided by the Exclusion Order remains a significant regulatory advantage that could be strengthened by incorporating a more substantive inquiry into whether apparent retail sales reflect genuine consumer demand or merely facilitate recruitment-driven compensation. Such an approach would preserve the clarity of the existing statutory framework while more closely aligning regulatory enforcement with Parliament’s original objective of preventing economically unsustainable recruitment-driven schemes. XI.

RECOMMENDATIONS

A. Adopt an economic reality assessment Singapore should supplement its existing statutory criteria with an economic reality assessment similar to that developed in the U.S. Rather than focusing solely on whether compensation is formally linked to product sales, regulators should also examine whether compensation is ultimately supported by genuine retail demand or by apparent sales driven primarily by recruitment-based purchasing. Some relevant considerations may include the proportion of sales made to ultimate users, the extent of internal consumption, qualification purchases, and whether the business could remain commercially viable without continual network expansion. This approach would better distinguish legitimate direct-selling enterprises from PPS that masquerade as MLMs and merely satisfy the Exclusion Order's formal requirements while remaining economically dependent on recruitment. B. Introduce mandatory earnings disclosure requirements Singapore should also introduce mandatory earnings disclosure requirements comparable to the FTC. Presently, Section 2(1)(c)(v) of the Exclusion Order merely requires promoters to maintain fair and accurate audited records relating to benefits received by participants without disclosure to them. Before joining, prospective participants should instead receive standardised disclosures that provide meaningful data on distributor earnings, including median and percentile earnings, the proportion of participants earning a net profit after business expenses, participant attrition rates, and the distribution of earnings across the network. Such disclosures would provide a more realistic picture of likely financial outcomes, reduce misleading 346

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

optimism created by exceptional success stories and enable prospective participants to make informed commercial decisions. C. Introduce licensing and competency requirements for recruiting distributors Singapore should reconsider whether recruitment in MLMs should remain effectively unregulated. Unlike franchisees, insurance representatives and real estate agents, MLM participants may recruit members of the public, make earnings representations and promote commercial opportunities without any minimum competency requirements, licensing obligations or regulatory supervision. Franchise arrangements ordinarily involve significant due diligence before appointment, including applications, interviews, capital assessments, business planning and operational training. Similarly, insurance representatives and estate agents are subject to mandatory examinations, 101 licensing requirements, continuing professional education and disciplinary oversight because they deal directly with members of the public in circumstances involving information asymmetry and financial vulnerability. By contrast, MLM recruitment frequently requires little more than the purchase of an initial product package or starter kit. Individuals with no business experience, sales training or understanding of compensation structures may immediately begin recruiting others and making representations regarding earnings potential. The very accessibility that enables rapid network expansion also increases the likelihood that inexperienced and financially vulnerable individuals become both recruiters and recruits. Therefore, Singapore should consider introducing a similar licensing regime for MLM distributors. Such individuals could be required to complete training and be given a registration number subject to continuing regulatory oversight. The foregoing recommendations do not require Parliament to prohibit all MLM arrangements. The rationale underlying the Exclusion Order remains understandable. Legitimate directselling businesses may provide genuine entrepreneurial opportunities, and Singapore has consistently preferred proportionate regulation over blanket prohibitions. This approach reflects Singapore’s broader philosophy of economic regulation. As Minister Chee Hong Tat

101

Institute of Banking and Finance Singapore, Capital Markets and Financial Advisory Services (CMFAS) Examinations https://www.ibf.org.sg/home/exams/examinations/cmfas/examination-details/register accessed 14 July 2026; Council for Estate Agencies, Prepare for the Real Estate Salesperson Examination (2 June 2026) https://www.cea.gov.sg/real-estate-professionals/for-aspiring-real-estate-salespersons/real-estate-salespersonexamination/prepare-for-the-real-estate-salesperson-examination/ accessed 14 July 2026.

347

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

recently observed in the context of wealth management, Singapore must be prepared to accept a degree of regulatory risk where doing so enables economically valuable industries to flourish.102 The objective is therefore not to eliminate every rogue actor, but to calibrate regulation so that the economic benefits of an industry justify the residual risks that remain despite effective safeguards and enforcement. That policy rationale is compelling in sectors such as wealth management and financial services, where innovation, capital formation and employment generate substantial public benefits notwithstanding occasional misconduct. The analysis in this article suggests that the same policy calculus is considerably less persuasive in relation to MLMs. Although Singapore generated approximately US$591 million in annual direct-selling sales,103 those figures do not distinguish genuine retail demand from purchases motivated principally by qualification requirements, internal consumption, or recruitment incentives. More fundamentally, the evidence considered throughout this article suggests PPS can comply with formal requirements and operate lawfully despite retaining the objectionable features Parliament originally associated with pyramid selling. It does not appear to be an isolated instance of regulatory noncompliance but rather a recurring structural feature of compensation systems that remain economically dependent on continual network expansion. Parliament should therefore place a greater evidential burden upon promoters to demonstrate that their schemes are supported by genuine retail demand rather than recruitment-driven purchasing. Doing so would preserve legitimate direct-selling enterprises while ensuring that statutory compliance reflects commercial substance rather than legal form, thereby better aligning Singapore’s regulatory framework with Parliament’s longstanding objective of protecting consumers from the substantive harms associated with pyramid selling. XII.

CONCLUSION

Singapore’s legislative framework has largely succeeded in achieving the objective Parliament set for it in 2000: distinguishing lawful MLM arrangements from unlawful pyramid selling

102

Sheila Chiang, ‘Singapore has to accept risk of rogue actors to prosper as a wealth hub: Chee Hong Tat’ The Straits Times (8 January 2026) https://www.straitstimes.com/business/spore-has-to-accept-risk-of-rogue-actorsto-prosper-as-a-wealth-hub-chee-hong-tat accessed 15 June 2026. 103 World Federation of Direct Selling Associations, WFDSA STATS: 2024 Global Annual Direct Selling Statistical Data Report (November 2025) 47 < https://wfdsa.org/wp-content/uploads/2025/12/WFDSA-STATSReport-2024-2025-V1.pdf> accessed 15 June 2026

348

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

through clear statutory criteria centred upon the source of participant remuneration. The Exclusion Order provides commercial certainty, accommodates legitimate direct-selling businesses and avoids unnecessarily prohibiting entrepreneurial activity merely because it adopts a multi-level compensation structure. Judged by its own formal criteria, the legislative framework is coherent and internally consistent. This article has argued, however, that the central question is not whether the Exclusion Order operates coherently as a matter of legal form, but whether it continues to achieve Parliament’s broader objective of protecting consumers from the substantive harms historically associated with pyramid selling. Examining the legislation through the PPS framework, empirical research, investigative journalism and participant accounts demonstrates a growing disconnect between formal statutory compliance and commercial reality. Contemporary MLM compensation structures increasingly monetise recruitment indirectly through qualification purchases, internal consumption, team-volume incentives, and expanding distributor networks, enabling schemes to satisfy statutory safeguards while preserving the same economic dependence on continual recruitment that characterises traditional pyramid schemes. The widespread financial losses, inventory loading, exaggerated earnings expectations, misleading product claims and social pressures documented throughout this article therefore do not merely reveal isolated enforcement failures; they suggest structural limitations within the legislative framework itself. The comparative analysis further demonstrates that a different methodology is possible. Rather than treating product sales as determinative, the U.S. asks whether those sales reflect genuine consumer demand or merely facilitate recruitment-driven compensation. Although the U.S. approach has its limitations, its approach more directly addresses the underlying commercial reality of MLM operations. Nevertheless, Singapore does not need to transplant the American framework completely, but it should reconsider whether its present statutory inquiry is best positioned to identify the very schemes Parliament originally sought to eliminate. This article therefore questions not merely whether the Exclusion Order should be strengthened, but whether MLM arrangements should continue to receive a categorical statutory exclusion at all. Throughout the legislative debates, Parliament consistently recognised that some degree of regulatory risk was justified to accommodate legitimate commercial activity. That policy calculus remains sensible in sectors where economic 349

Singapore Comparative Law Review

“How Is This Legal?”: Reassessing Multi-Level Marketing Regulation In Singapore

contribution clearly outweighs the risks posed by a few bad actors. However, the evidence surrounding MLMs presents a markedly different picture. Despite decades of statutory safeguards, studies continue to report overwhelming evidence of participant losses, investigative journalism repeatedly documents misleading promotional practices, and public perception remains overwhelmingly characterised by distrust and controversy. At the same time, the industry’s demonstrated economic contribution to Singapore appears comparatively modest, and existing sales figures do not distinguish genuine retail demand from purchases driven by qualification requirements or internal consumption. Therefore, the continued operation of MLMs in Singapore requires a stronger justification than the present framework appears able to provide. Ultimately, the issue is whether preserving MLM as a lawful business model is worth the risks it poses to participants and society. The evidence considered throughout this article suggests that it does not. Where an industry consistently produces widespread participant losses, recruitment dependence, misleading promotion, and persistent reputational concerns, notwithstanding formal legal compliance, the burden should not remain on consumers to distinguish legitimate opportunities from sophisticated PPS. Parliament itself recognised the complexity of distinguishing legitimate direct-selling arrangements from prohibited pyramid schemes when it enacted the Exclusion Order and its detailed statutory safeguards. If legislators and regulators continue to grapple with that distinction, it is difficult to expect the public to perform the same legal and economic assessment before joining an MLM. Yet the current regulatory position effectively places that responsibility upon the public by requiring them to exercise vigilance104 while sophisticated operators continue to innovate around statutory safeguards. The burden should instead fall on the industry to justify its continued statutory exclusion from legislation enacted precisely to prevent these harms. If the predominant evidence is that modern MLMs have evolved beyond the assumptions upon which the Exclusion Order was founded, then the law should evolve with them.

104

Ministry of Trade and Industry, ‘Multi-level Marketing and Pyramid Selling Act’ (12 December 2025) FAQ Q.10 https://www.mti.gov.sg/resources/laws-and-regulations/multi-level-marketing-and-pyramid-selling-act/ accessed 15 June 2026.

350

Headquartered in Singapore, WongPartnership is one of the country’s largest law firms, offering a comprehensive range of legal services with recognised depth in advisory, disputes, regulatory and transactional work. Through its regional law network, WPG, WongPartnership provides integrated legal support across Asia and the Middle East. The network includes Al Aidarous in Abu Dhabi and Dubai, Makes & Partners in Indonesia, Cheang & Ariff and Foong & Partners in Malaysia and Gruba Law in the Philippines. The Firm is consistently recognised as a Top-Tier Firm by established legal publications including Chambers, IFLR1000 and The Legal 500.

An extraordinary opportunity awaits. Your future starts here. As one of Singapore’s top-tier law firms, we’re committed to developing future leaders in the legal profession. Through personalised mentorship, rigorous training, and real-world experience, we help our lawyers reach their full potential. We are accepting applications for rising third-year law undergraduates from scheduled overseas universities who wish to undertake relevant legal training with our firm in 2027.

Scan and learn more about our Training Contract and Internship Programme!

For a closer look at our programme and how to apply, download our trainee recruitment brochure. If you're ready to take the next step in an extraordinary legal career, we’d love to hear from you. Connect with us on LinkedIn here.

ASEAN | CHINA | MIDDLE EAST WongPartnership LLP (UEN: T08LL0003B) is a limited liability law partnership registered in Singapore under the Limited Liability Partnerships Act 2005.

Scan and stay in touch with us!