The AI Agent Security Field Guide
The AI Agent Security
Field Guide
The OWASP Agentic Top 10 mapped to real attack patterns — with the mitigations your team can deploy immediately.
Free · No second form · Immediate download
All 10 ASI risks mapped
Each OWASP Agentic Top 10 entry explained with a real incident reference, attack mechanism, and 2–3 specific controls.
Production-focused
Written for teams running agents now — not theoretical. Every control maps to something you can implement this week.
Real incidents, named
DockerDash, WhatsApp MCP, CVE-2025-6514, Gemini memory corruption — the actual breaches, not hypotheticals.
Actionable mitigations
mcp-scan, tool description hashing, human approval gates, per-server namespacing — specific tools and techniques.
What’s inside
All ten OWASP Agentic risks, ASI01 through ASI10. For each one: an incident where it actually happened, how the attack mechanism works, and two or three controls you can put in place.
The MCP sections go deeper, into tool poisoning, meta-context injection, and cross-server attacks, with code for each. Incidents are named rather than anonymised — DockerDash, the WhatsApp MCP exfiltration, CVE-2025-6514, the Gemini memory corruption — so you can go and read the postmortems yourself.
Controls are specific enough to act on the same week: mcp-scan, tool description hashing, human approval gates, least-privilege architecture, per-server namespacing.
I wrote it for security engineers and CTOs who already have agents running in production.
Free · No second form · Immediate download
Also: the course this guide was written for → Join the waitlist
Amine Raji, PhD · CISSP · aminrj.com