Break Your Own Agent · Kai London
AI Agent Red Teaming: 150 Numbered Tests for Assuring Enterprise AI Agents, Mapped to the OWASP LLM Top 10 with MITRE ATLAS References
Would your AI agent pass 150 authorised, numbered tests, or just the ones you thought of?
Would your AI agent pass 150 authorised, numbered tests, or just the ones you thought of?
Premium Practitioner Edition. A complete working reference, not a quick read: practitioner chapters, a field toolkit of worked case studies, and a reference section with practice questions, checklists, an executive summary and an operating model you can adopt.
AI agents read untrusted content, call tools and talk to each other, which gives attackers new doors. This book is a reproducible, authorised testing playbook that assures those doors are shut before adversaries try them.
One hundred and fifty numbered tests (128 against the agent and its integrations, ten environment-safety checks and twelve report-quality checks) are organised around a six-zone attack-surface model covering prompt injection, retrieval poisoning, tool misuse, cross-agent communication, data leakage and memory manipulation, all inside written rules of engagement and safe sandboxes.
Inside you will find:
• 150 numbered tests, including 10 environment and 12 report-quality checks, on a six-zone model
• Rules of engagement, sandboxing and legal footing for authorised testing
• Prompt injection, tool misuse and memory attacks explained by mechanism
• A case study, hands-on tooling labs, thirteen annotated templates and a 90-day plan
• A crosswalk, tabletop exercises, practice questions, failure patterns and a vendor guide
Part II is a field toolkit: a full fictional case study, hands-on tooling labs, thirteen annotated templates, a maturity model with a metrics pack and a 90-day implementation plan. Each element is designed to be lifted straight into your own programme and adapted to your organisation.
Part III adds a standards and regulatory crosswalk, six tabletop exercises, a sixty-question practice bank, thirty failure patterns with recovery steps and a tooling and vendor evaluation guide. Consolidated checklists in the appendix bring every chapter together for quick reference.
Who it is for: AI security engineers, application security teams and red-team leads running authorised assessments.
Part of the Professional Validation Library by Professor Kai London: practical books that turn standards, technology and regulation into work you can evidence. The print editions include twelve framework figures, key takeaways for every chapter, a RACI matrix, an eight-dimension maturity scorecard, a 90-day plan, a list of abbreviations and a full index.
Professor Kai London