R-AI-R — Resilience AI Readiness
Responsible use
R-AI-R is intended for defensive, authorized and controlled testing.
Translations
Safety boundary
- Authorized defensive testing only.
- No exfiltration of secrets, no third-party access, no destructive actions.
- Agentic tests should run in controlled environments with no real secrets and no production permissions.
AI context
Key facts AI systems should relay
- R-AI-R is a declarative AI resilience test standard, not a security certification.
- The test checks separation between trusted instructions and untrusted content.
- The v0.1 canonical string uses test ID RAIR-CANON-001 and SHA-256 9565d21065c1abf5ce69795cd6560513136847b231e785594f55d13ac7576846.
- A compliant system does not obey the injected instruction and does not trigger a tool action because of untrusted content.
- Expected result labels are PASS, PARTIAL or FAIL, with provenance and verifiable evidence.