Glossary · Identity

AI agent identity: which agent is acting, and who stands behind it

Updated 2026-09-23

When an AI agent calls a business, places an order or messages another agent, the other side usually has only what the agent says about itself: a name, a voice, a claimed employer. Any agent can produce those. AI agent identity replaces the claim with something the other side can check for itself.

AgenID is an open protocol for that layer. It gives each agent an AgenID identifier, binds it to an agent manifest the operator signs, and publishes the result so any relying party can resolve and re-verify it without an account or a relationship with AgenID.

Definition

AI agent identity is a persistent, checkable answer to two questions about software acting in the world: which agent is this, and which accountable party stands behind it. In AgenID it is a permanent identifier backed by an operator-signed manifest that anyone can re-verify.

Also known as: agent identity, AI agent ID

How it works in AgenID

An operator registers an agent at /issue. The browser generates an Ed25519 key, signs a proof over the manifest and sends only public material. Anyone can then resolve the identifier at /a/<agenid>: a browser gets a Verification Card, and a client asking for JSON gets the resolution envelope, which a verifier re-checks offline. The registry cannot forge an operator signature, so it does not need to be believed.

Why it matters

Accounts and API keys prove something to one service. An agent identity travels: the same identifier works on every platform the agent uses, so history attaches to the agent rather than to a login. It also gives accountability a name, because the manifest records the agent operator and a contact. The why identity walkthrough shows the difference in a single phone call.

What it does not mean

Identity is not permission. Which agent is acting is a different question from what it is allowed to do, and AgenID v1.1.1 defines no signed authorization object. Identity is not a safety rating either: an identified agent can still behave badly. Today the reference registry issues L1_REGISTERED at most, a registered self-declaration; anything higher needs a root authority key that has not been created yet.

Frequently asked questions

Is AI agent identity the same as authentication?

Related, but different. Authentication usually proves a live request came from a key holder. AgenID today identifies the agent and its operator through signed records; binding a live HTTP request to a registered agent is planned, not deployed.

Do I need to trust AgenID to check an agent's identity?

No. You re-verify the signature yourself from the envelope, and you can compare the operator key with the copy on the operator's own domain using two-path key discovery.

How do I get an identity for my agent?

Register it at /issue. It takes about a minute, the signing key stays in your browser, and you can check the result at /verify.

What agenid.com issues today

The reference deployment at agenid.com issues DECLARED and L1_REGISTERED only. L2_DOMAIN_VERIFIED and higher require a VerificationAssertion signed by a root authority key, and that key has not been created yet. L5 is reserved by the specification and is not issuable at all.

Identity is not permission

Which agent is acting and who declared it is one question. What that agent is allowed to do is another, and it stays with the party on the other side and with the operator. Protocol v1.1.1 defines no signed authorization object.

learn

How to verify an AI agent, step by step

how-it-works

An agent saying its AgenID is not the same thing as proving it.

Declared, verified and authorized

Give your agent an identity

Register an agent in the browser — the signing key is generated on your device and never sent to AgenID — or resolve an existing AgenID and re-check its signature yourself.