AI Red Teaming and Offensive AI Security All-in-One: The Professional Handbook for Prompt Injection, Agent Exploitation, LLM Attacks, RAG Poisoning, ... Abuse, Adversarial Testing and AI Defence

AI Red Teaming and Offensive AI Security All-in-One is an 805-page professional handbook for prompt injection, agent exploitation, LLM attacks, RAG poisoning, MCP security, model abuse, adversarial testing and AI defence. Organised around the Adversarial Cycle, it treats the testing of AI systems as a board obligation with a defined threat model, governing standards and a repeatable programme, not a one-off stunt.

Part I sets out why testing AI systems is now a board obligation, the threat model in terms of positions, goals and techniques, and the standards and authorities that govern the work. Part II covers scoping and discovering the targets: building the AI inventory, finding every model, app and agent and the blind spots of discovery, classifying by exposure, reachability and blast radius, and third parties, SaaS and the AI supply chain. Part III turns to prioritisation, programme design and the business case.

Later parts walk through the offensive techniques themselves, from direct and indirect prompt injection, jailbreaks and system-prompt extraction to tool and agent exploitation, retrieval and RAG poisoning, MCP and tool-calling abuse, model theft and abuse, and adversarial machine learning, and then the defences, detection, evidence and governance that turn findings into durable controls.

Written for red teamers, penetration testers, AI security engineers, security architects, SOC leaders, CISOs, auditors and governance professionals responsible for AI systems in production. Part of the Understand, Architect, Attack, Defend, Govern series of enterprise AI security references.

Key topics include: AI red teaming; prompt injection; jailbreaks; agent exploitation; RAG poisoning; MCP and tool-calling security; model abuse and theft; adversarial machine learning; AI inventory and discovery; AI supply chain; AI security programme design; AI defence and governance.

Professor Kai London is a cybersecurity executive, security architect, author and professional researcher with extensive experience across cybersecurity, critical infrastructure, resilience, governance, risk and technology. His work focuses on complex operational, security and institutional challenges, including cyber resilience, infrastructure protection, evidence, governance, recovery and reconstruction. He writes practitioner-focused professional reference works designed for executives, specialists, researchers and institutional readers.