June 4, 2026
Open-source methodology by OWASP, CoSAI, CSA, and NIST contributors enables enterprises to compare AI agents by security risk for the first time
Read moreJune 4, 2026
Open-source methodology by OWASP, CoSAI, CSA, and NIST contributors enables enterprises to compare AI agents by security risk for the first time
Read moreJune 4, 2026
Open-source methodology by OWASP, CoSAI, CSA, and NIST contributors enables enterprises to compare AI agents by security risk for the first time
Read more
The full technical reference behind our ASI05 guide — five attack vectors in depth, a five-layer defense architecture, and an architect's checklist.
Our definitive guide to OWASP ASI05 provides the high-level taxonomy and technical definitions of unexpected code execution in agentic AI. Building a secure agentic architecture, however, requires detailed mitigation strategies — that is what this whitepaper delivers.
When an AI agent has the autonomy to write and run code, the barrier between a natural-language prompt and remote code execution evaporates. This reference takes each of the five ASI05 attack vectors from taxonomy to defense: concrete controls, an architecture framework that layers them, and a checklist to audit an existing agentic stack against.
Security professionals, architects, and risk managers responsible for agentic AI systems that generate or execute code — anywhere a prompt could become an instruction to your infrastructure.
OWASP ASI05 — Unexpected Code Execution: Defense Architecture Technical Reference
March 9, 2026
