AI Security WikiEvidence before assurance.

Finding record

Save actual assessment records in the caller’s private assessment workspace, not this public repository.

  • Title / ID:
  • Target and authorized scope:
  • System version, environment and timestamp:
  • Deployment patterns and rationale:
  • DASF risks, controls and source snapshot:
  • Status: unverified / confirmed / refuted / resolved with retest evidence
  • Asset, actor and trust boundary:
  • Expected behavior:
  • Observed behavior:
  • Evidence references: artifact path or approved locator, relevant excerpt, timestamp; redact secrets
  • Coverage and verification: coverage unit, discovery identity, separate verifier if used, strongest existing control and exact unresolved facts; do not imply an independent review that did not occur
  • Reproduction and legitimate comparison: exact authorized synthetic inputs and outcomes
  • Impact and reachability: supported facts, uncertainty and stakeholder assessment
  • Correction proposal: enforcement location, minimal change, dependencies, owner
  • Rollback: restoration procedure and trigger
  • Verification: prohibited operation is blocked, legitimate operation works, attributable logs exist
  • Residual risk / owner decision: no acceptance may be inferred from silence or passing tests
  • Explanation for the user: mechanism → concrete impact → correction → verification, in requested language