Content-derived
The agent ID is derived from provider, platform, runtime and agent attributes, deterministic and stable.
Identity
A persistent identity for every AI agent.
Agent Identity gives each AI agent a stable, content-derived identity, the same agent resolves to the same ID across executions, so trust, passport and reputation attach to a real, persistent subject.
Definition
What an agent identity is, and what it is not.
A subject that evidence can be attached to. Nothing more, and deliberately nothing more.
Registry-backed
Identities are coordinated on a persistence store via the agent registry.
Not auth
Identity is a trust subject, not an IAM account, it complements your existing auth.
Lifecycle
Observe once, resolve forever.
- Derive a stable agent_id from operation context
- Register and update the identity as it is observed
- Attach trust, passport and reputation to the same subject
POST /api/v1/registry/observe GET /api/v1/registry/agents/{id} { "agent_id": "agent_9f2c...", "operation_count": 128, "latest_trust_score": 82 }
Boundaries
Identity, evidence and the Passport are three things.
They are routinely collapsed into one word, and collapsing them is how a trust document gets mistaken for a login. The Agent Passport page is the canonical definition of the passport; this page defines the subject it is about.
- Agent Identity
- The subject. A stable, content-derived
agent_idthat the same agent resolves to across executions, so a record made today and a record made next quarter are about the same thing. - Trust evidence
- The records. Sealed envelopes and receipts produced by the Evidence Engine, each citing the identity it was about.
- Agent Passport
- The document. A portable trust document composed from those records. It is a trust document rather than a credential, it is not self-declared, and it makes no claim of external standards interoperability.
- Authentication
- Not this. Proving who is calling is your identity provider’s job, and VerifAIer integrates with none: there is no OAuth, no OIDC and no SSO.
- Authorization and IAM
- Not this either. An
agent_idgrants nothing and is never checked before an action. Access control inside the product is organization membership and roles, which is a different mechanism with a different name. - Runtime governance
- Deciding what an agent may do while it runs is Sentinel Runtime’s job. Identity says who acted; it never says who may act.
Note
The trust score in the example above is scoped to one agent and derived from that agent’s own observations. There is no estate-wide composite score, and nothing here aggregates one.
One evidence layer
Related platform components
Not separate tools keeping separate records. Each of these derives from the same attested evidence, so you can adopt them in any order.
Give your agents an identity.
Persistent identity is the foundation for portable AI trust.