Skip to content

Security: MMVFIRM/Universal-Earned-Autonomy-Layer

Security

SECURITY.md

Security Policy

Reporting

Report vulnerabilities privately to security@mmv.example (placeholder). Do not open public issues for security reports.

Security properties this layer is designed to provide

  • Agent authentication: every workflow event is Ed25519-signed; identity is a public key, not a self-asserted string. Unsigned/forged events are rejected in strict mode.
  • Enforcement, not advice: state-changing actions execute only through a Policy Enforcement Point that verifies a capability token against the actual action. A blocked/approval decision issues no token.
  • Replay resistance: per-agent event nonces and idempotent, one-per-event decisions prevent replay-based trust inflation.
  • Risk ceilings: high-risk/irreversible authority classes cannot reach autonomous execution regardless of trust history.
  • Tamper-evident audit: hash-chained, Ed25519-signed records with periodic signed checkpoints and an external-anchor hook.

Properties that depend on correct deployment (see docs/THREAT_MODEL.md)

  • Private key custody (HSM/KMS) for issuer and agent keys.
  • A real OIDC IdP for human/admin authentication.
  • External anchoring of ledger checkpoints to immutable storage.
  • Network isolation of the PEP from agents (the PEP must be the only path to downstream side effects).

There aren't any published security advisories