Applied AI Security: From Models to Agents

Artificial intelligence has stopped being something you talk to and started being something that acts. The moment a model can send an email, call an API, run code or move money, every weakness in it becomes a path to real-world consequences. Applied AI Security: From Models to Agents is the single, coherent guide to defending these systems, written for the people who now have to. This is a hands-on, lab-driven textbook that takes you from the statistics of a single classifier to the security of autonomous, tool-using agents. You will not just read about attacks; you will build them, then build the defences, against a target system that runs throughout the book. What you will learn - The AI attack surface, threat modelling with MITRE ATLAS, and the machine-learning supply chain. - Attacks on models: adversarial examples, data poisoning and backdoors, model extraction and theft, and inference and privacy attacks, with the privacy-preserving defences that counter them. - Securing LLMs and agents: prompt injection, jailbreaks and misuse, RAG and vector-store security, agent sandboxing, the tool, skill and Model Context Protocol supply chain, and multimodal and deepfake threats. - Assurance and governance: red-teaming and evaluation, detection, monitoring and incident response, AI governance, risk and compliance (NIST AI RMF, ISO/IEC 42001, the EU AI Act), and a full end-to-end capstone. Why this book is different - Current to 2026. Every framework version, CVE and incident is verified against primary sources and dated, so you can see exactly how fresh each claim is. - Lab-driven. Twenty hands-on labs, one per chapter, run offline on a laptop with no paid API key. The full notebooks, Docker environments and source live at the companion site. - Built for two readers. Security professionals learning the new material, and students from undergraduate to MSc. It assumes you know security and teaches you the machine learning you need. Who it is for Penetration testers, security engineers, SOC analysts, red-teamers, architects and students who have just been handed a system with a model in it and need a place to start. A companion to Applied Cybersecurity: Foundations to Mastery, this volume goes deep on the one thing that changes everything: securing systems whose behaviour is learned, not written. About the author Dr Matt Lemon is a CISO with more than twenty-five years in cybersecurity and the founder of ShieldIQ. He teaches this material and secures real systems, and both feed the book.