OWASP AISVS (AI Security Verification Standard)

A versioned OWASP standard (1.0 released June 2026, modeled on OWASP ASVS's locked-version-per-release policy) for verifying AI-system security, with a control family — C9 "Orchestration & Agentic Security" — built specifically for autonomous and multi-agent systems. Matters to this vault as the first standards-body text checked directly against the vault's TOCTOU-in-agents thread: as of v1.0 it builds an entire chapter around the surface where a stale plan meets a changed world, yet never names TOCTOU, race condition, atomicity, or lost update, and its closest analog (9.5.6) re-checks authorization freshness rather than world-state freshness. A live document — OWASP maintains a 1.01-dev folder for the next minor version — so this page will likely need revisiting as the standard evolves. As of 2026-07-26, 1.01-dev mirrors v1.0's C9 structure exactly and still carries no TOCTOU/atomicity/concurrent-agent language, and no release newer than v1.0 has shipped. Project leads include Jim Manico, Otto Sulin, Rico Komenda, Russ Memisyazici, and Raza Sharif.

References

• claim-aisvs-1-0-omits-toctou-and-lost-update-terminology
• claim-aisvs-closest-analog-to-toctou-is-authorization-freshness-not-world-state-recheck
• claim-bollen-reports-unverified-aisvs-concurrent-agent-toctou-requirement
• claim-aisvs-101-dev-branch-omits-toctou-as-of-2026-07-26
• observation-webfetch-fabricated-aisvs-issue-640-pr-641-across-two-sessions
• question-verify-bollen-aisvs-concurrent-toctou-requirement