Platform

Agent identity

An agent is not a user and not a service account. It needs an identity of its own that can be scoped, rotated and revoked without touching the systems it acts on.

How identity works

  • A UUID identifies the agent, minted at registration.
  • Provisioning generates 32 random bytes, hex-encoded, and returns the raw value once.
  • What is stored is a salted PBKDF2 hash with a pepper, never the secret itself.
  • Verification uses a constant-time comparison, so a rejected credential leaks no timing information.
  • Rotation replaces the stored hash immediately.
  • Revocation flips the agent’s state so the credential stops verifying.

Session tokens carry extended claims — trust domain, blast radius, session id and execution context — so a downstream service can decide whether to honour a call without re-reading the registry.

What identity is not, yet AWARE’s identity layer authenticates the holder of a shared secret. It does not bind an identity to a key that only the agent holds. There is no X.509 certificate, no Ed25519 key pair and no SPIFFE identifier in the current implementation. Describing it as “cryptographic identity that cannot be faked” would overstate it.
Why the difference matters A bearer secret can be replayed by anything that obtains it. A key-bound identity cannot. AWARE closes the replay gap with short-lived sessions, revocation and blast-radius scoring today; key binding is the natural next step.

Attestation

Agents present a signed session token. The attestation service verifies it, checks that the session is live, that the agent is still active, and that the trust domain matches the target’s.