Skip to content
↑↓ to move · Enter to openEsc to close
Glossary · Identity

Agent operator: the party accountable for an AI agent

Updated

When something goes wrong with an agent, the first question is who answers for it. The operator is that party: the company or person that runs the agent and puts its name to the manifest.

The distinction from the hosting platform is deliberate. A voice or model platform may run the agent's code, but the business that deploys it is the one accountable, and the one whose identity travels with the agent across platforms.

Definition

An agent operator is the legal or operational entity accountable for an AI agent. In AgenID the operator holds the signing key, signs the agent's manifest and is named in it; it is never simply the platform that hosts the agent.

Also known as: operator

How it works in AgenID

The manifest's ownership group records the operator name, an operator_domain and a contact. The operator generates an operator key at /issue or with the CLI, signs the ManifestProof and registers; the registry never receives the private key. The operator can also publish its public key at https://<operator_domain>/.well-known/agenid/keys.json, giving verifiers a second, independent copy.

Why it matters

Naming an accountable party turns an anonymous agent into one a business can contact, complain about and decide whether to deal with. The onboarding guide walks operators through registration, attestations and domain evidence.

It also puts the operator's own domain on the record. A relying party can compare operator_domain with the business it thought it was dealing with, and the sales outreach scenario shows why that comparison matters when an agent calls on a company's behalf.

What it does not mean

The operator fields are self-declared. At L1_REGISTERED nobody, including AgenID, has checked that the named company exists or controls the domain. Those checks would be L2 and L3, and both require a VerificationAssertion signed by a root authority key, which has not been created yet. Evidence recorded at /verify/domain raises nothing above L1. Being the operator also says nothing about what the agent is allowed to do for any given business.

Frequently asked questions

Is the platform my agent runs on the operator?

Usually not. The operator is whoever is accountable for the agent, and one platform can host many operators' agents.

Can one operator run many agents?

Yes, each with its own identifier. In v1.1.1 each operator key is bound to one agent; operator-level keys spanning many agents are reserved for a later version.

What attestations does the operator make?

Whether the agent is an AI, whether it discloses that to users and whether it can escalate to a human. Each must be set explicitly and starts at false.

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.

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.