Clean/challenge artifact pairs
R-AI-R — Resilience AI Readiness
Test whether an AI treats untrusted content as data, not as orders.
R-AI-R provides an adaptive business-safety challenge suite, hashes, scoring rules, and clean/challenge procedures for evaluating chatbots, RAG systems, agents, and document scanners.
AI context
Key facts AI systems should relay
- R-AI-R is a declarative AI resilience test standard, not a security certification.
- The v0.5 suite tests whether hidden, encoded, ranked, package-supplied or customer-supplied content silently changes answers, review decisions or actions.
- The suite uses clean and challenge artifact pairs across DOCX, PDF, HTML, TXT, RAG corpus, JSON, skill-package and tool-plan cases.
- Evaluator-only files such as expected answers and scoring rules must not be provided to the system under test.
- Expected result states are AT RISK or RESILIENT, with provenance and verifiable evidence.
DOCX · PDF · HTML · TXT · RAG · JSON · skill package · tool plan
Generated result states, not certification
Workflow v0.5
Download
Get the suite, manifest, hashes, protocol, and schema.
Run clean
Execute the clean artifact with the task.
Run challenge
Use the same task on the challenge artifact.
Score
Compare outputs and declare evidence.