International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 11 | Nov 2025
p-ISSN: 2395-0072
www.irjet.net
AI and ML in Penetration Testing: A Comprehensive Review Soham Deshmukh1, Mohanish Kulkarni2 , Jayam Mehta3 , Pallavi Akolkar4 1,2,3,4,Department of Information Technology,
Trinity College of Engineering and Research, Pune, India ---------------------------------------------------------------------***--------------------------------------------------------------------2.BACKGROUND AND THEORETICAL FOUNDATIONS Abstract - This review surveys state-of-the-art research on the integration of Artificial Intelligence (AI) and Machine Learning (ML) in cybersecurity, critically synthesizing findings across 20+ foundational papers. Spanning both defensive and offensive perspectives, the analysis covers automated penetration test ing, deep learning-based intrusion detection, explainable AI, adversarial attacks, IoT and industrial infrastructure security, and educational strategies for developing future expertise. Key results highlight substantial progress in automating vulnerability identification, scaling threat prediction, and enhancing attack simulation. Limitations persist in adversarial robustness, explain ability, real-time adaptation, and human-AI collaboration. The literature lacks standardized performance metrics and sufficient cross-sector partnerships. The paper synthesizes consensus and debates, identifies critical gaps, and lays out future research direc tions that prioritize ethical, scalable, and resilient cybersecurity solutions leveraging synergistic AIhuman teams.
2.1 Foundational Concepts AI encompasses a suite of intelligent computational methods, including supervised, unsupervised, and reinforcement learning, with applications extending to deep learning (DL) architectures such as CNN, RNN, LSTM, GAN, and DBN. ML models feature hierarchical feature extraction and data driven learning that optimize for both detection accuracy and adaptability. Key concepts include adversarial machine learning— attackers manipulating training data or models, explainability—making AI model decisions interpretable to humans, and automated planning—using AI agents to simulate real-world attack and defense actions.
2.2 Standardized Frameworks Research increasingly builds on operational and adversarial frameworks: • MITRE ATT&CK and CAPEC: Attack taxonomy and procedural modeling [19], [20]. • Lockheed Martin Cyber Kill Chain: Structured phases of cyber attacks. • OWASP IoT Framework: Definitions of IoT attack surfaces. These frameworks anchor experimental setups, vulnerability assessments, and the training of automated AI/ML security agents.
Key Words: Artificial Intelligence; Machine Learning; Cy bersecurity; Penetration Testing; Deep Learning; Explainable AI; Adversarial Attacks; Intrusion Detection
1.INTRODUCTION Cybersecurity has ascended to strategic prominence in an era of ubiquitous connectivity, smart infrastructure, and digital transformation. Network and application attack surfaces have grown, and manual security processes now struggle to contend with the scale and sophistication of cyber threats. The emer gence of Artificial Intelligence (AI) and Machine Learning (ML) offers a paradigm shift, enabling automated, adaptive, and predictive mechanisms for threat detection, mitigation, and security operations management. This review explores the recent literature defining the most impactful advances, the consensus and controversy surround ing major approaches, and unaddressed obstacles that impede robust, ethical, and effective deployment. The objectives are to evaluate core methodologies, unify perspectives from defense and offense, and provide a roadmap for future research and practice linking technical innovation with societal resilience.
2.3 Key Cybersecurity Activities The research papers target penetration testing (PT), intrusion detection systems (IDS), malware analysis, vulnerability scan ning, risk propagation modeling, adversarial threat simulation, and the development of cyber defense curricula.
3. LITERATURE REVIEW 3.1 Automated Penetration Testing and Attack Simulation:
Recent advances demonstrate AI-driven penetration testing frameworks, where reinforcement learning (often Deep Q E. AI for Critical Infrastructure and IoT Networks) and AI planners learn optimal attack paths in simulated network environments [4], [5], [7]. These systems mimic skilled
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 271