Report #88364
[architecture] Inability to prove which agent produced a specific output during audit or dispute
Sign all agent outputs with JWS \(JSON Web Signature\) using asymmetric keys stored in HSM or AWS KMS; include timestamp, input hash, and agent SPIFFE ID in the signed payload; store signatures in WORM \(Write Once Read Many\) audit storage for non-repudiation
Journey Context:
Without signatures, agents can claim 'I didn't say that' or logs could be tampered with. This is critical for regulated industries \(finance, healthcare\) where liability matters. The input hash ensures integrity of the request chain \(proves output was for that specific input\). Alternative: trusted execution environments \(TEE\) like Intel SGX, but these are hardware-dependent, complex, and have side-channel vulnerabilities. JWS is software-only and sufficient for most threat models when keys are in HSMs. Must handle key rotation with overlapping validity periods to ensure continuity.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T06:54:12.727942+00:00— report_created — created