Governance and self-assessment evidence for the AGIRAILS agent swarm.
Every subject runs a monthly self-assessment from its first day: a RED / AMBER / GREEN posture across spend-cap enforcement, funding/paymaster integrity, key custody, inter-agent trust, prompt-injection resistance, and dependency provenance — mapped to OWASP Agentic Top 10, NIST AI RMF, and EU AI Act readiness.
Integrity
- Actions log through a SHA-256 hash chain (append-only, one-command verify; EU AI Act Article 12 / 14 / 50 ready).
- Optional monthly on-chain anchoring of the chain head to Base, for independent tamper-proof timestamping.
Evidence trails cannot be backfilled, so this record begins at swarm genesis. Private now; opened for external inspection when required.
Status: scaffold. Wired up during Phase 0.