Adversarial Machine Learning: Mechanisms, Vulnerabilities, and Trustworthy AI Strategies

The book provides a comprehensive technical examination of how AI systems—predictive, generative, and agentic—can be attacked, manipulated, and exploited through various adversarial techniques including evasion, poisoning, privacy attacks, backdoors, prompt injection, and jailbreaking.

It covers the complete adversarial landscape across the AI development lifecycle, from training data poisoning and model backdoor insertion to inference-time attacks and exploitation of generative AI systems like large language models through prompt engineering and fine-tuning manipulation.

The structure progresses from foundational concepts (AI system anatomy, attack surfaces, adversary profiles) through specific attack methodologies (evasion, poisoning, privacy extraction) to cutting-edge threats against generative and agentic AI systems, including data leakage, hallucination exploitation, and autonomous threat loops.

Each chapter includes practical recommendations for defense and concludes with key concepts, suggesting the book balances technical depth with actionable security guidance for practitioners building or defending AI systems.

The content addresses multiple stakeholder perspectives—from security researchers and AI engineers to policymakers—reflecting the multifaceted nature of AI security challenges across white-box, black-box, and gray-box attack scenarios.