Securing AI Agents
$0.99
Buy on Amazon View on Apple BooksPrefer it in print? See every format on Amazon →
Price on Apple Books. Amazon’s price may differ.
- Published
Description
AI agents read email, open documents, browse the web, call business systems through tools and remember what they learned. Every one of those abilities is also a way in for an attacker. Securing AI Agents is a practical, defensive guide to building agents that stay safe when someone is actively trying to misuse them. Following one realistic project, an insurer's claims-processing agent that reads untrusted emails and documents, queries policy systems through Model Context Protocol tools and proposes payments for human approval, the book shows how to threat model agents with data-flow diagrams, trust boundaries and abuse cases; defend against direct and indirect prompt injection with isolation, privilege separation, dual-LLM and plan-then-execute designs; use guardrail models with measured thresholds; protect retrieval and memory from poisoning; design tools with least privilege, validation, idempotency and approval gates; give agents workload identity and scoped, delegated tokens; secure MCP servers; protect personal data across regions; sandbox code, browser and computer-use agents; and log, detect, respond, red team and gate releases. Controls are mapped to the OWASP Top 10 lists, MITRE ATLAS, the NIST AI RMF, ISO/IEC 42001 and the EU AI Act.
This book is an independent publication and is not authorized, sponsored or endorsed by any company whose products it mentions. All trademarks belong to their respective owners and are used only for identification. This book was drafted with generative AI tools and checked with automated tools. Tools, features and prices change often; confirm them in current documentation.