AI agent identity glossary

34 terms used when people talk about identifying, verifying and trusting AI agents — defined precisely, with the boundary of what each one does and does not mean. Where a term describes something the AgenID specification defines but agenid.com does not yet issue, the entry says so.

Identity

AgenID identifier: An AgenID identifier is the permanent, non-semantic name of an AI agent in the AgenID protocol, written agenid: followed by a 26-character ULID. It is assigned once, never reissued and never re-keyed when the agent changes platform.
Agent delegation: Agent delegation is the act of a person or organization, the principal, giving an AI agent authority to act on its behalf, usually within limits of scope, amount and time.
Agent manifest: An agent manifest is the operator's canonical-JSON self-declaration about an AI agent: its identity, its accountable operator, its purpose and how it discloses itself. In AgenID the manifest is never signed directly; a ManifestProof binds it by SHA-256 digest.
Agent operator: 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.
Agent provenance: Agent provenance is the verifiable record of where an AI agent comes from: which operator declared it, when, and with what stated purpose.
Agent-to-agent authentication: Agent-to-agent authentication is how one AI agent establishes which agent it is dealing with, and who operates it, before exchanging data or completing a task with it.
AI agent identity: 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.
Decentralized identifier (DID): A decentralized identifier (DID) is a W3C-standardized identifier of the form did:<method>:<method-specific-id> that its controller can create and prove control of cryptographically, without a central issuing authority.
Non-human identity (NHI): A non-human identity is any identity that belongs to software rather than a person, such as a service account, API key, workload credential or AI agent.
Relying party: A relying party is the service or person that depends on an identity claim to make a decision, such as a business deciding whether to continue a call with an AI agent.
ULID: A ULID (Universally Unique Lexicographically Sortable Identifier) is a 128-bit identifier made of a 48-bit millisecond timestamp and 80 bits of randomness, written as 26 Crockford base32 characters. AgenID uses ULIDs for agent and key identifiers.

Verification

Declared, verified and authorized: Declared, verified and authorized are AgenID's three claim states: what an operator says about its agent, what an authority has checked against evidence, and what the agent may do. None is ever inferred from another.
Domain control verification: Domain control verification is a check that the party making a claim can change DNS or web content for a domain, usually by publishing a challenge value in a TXT record or at a well-known URL.
Resolution envelope: The resolution envelope is the canonical JSON document AgenID returns for an agent identifier: the manifest, its operator-signed ManifestProof, per-assertion checks, the highest currently-valid level and key-discovery pointers, packaged so a verifier can re-check everything offline.
Verification assertion: A verification assertion is an authority-signed statement that one specific claim about an agent was checked against one piece of evidence, at one level, in one scope, for one time window, bound to one manifest version. It is the only object that produces VERIFIED.
Verification Card: A Verification Card is the human-readable page at /a/<agenid> that shows an agent's operator, declared purpose, signature check and current level, rendered from the same data as the JSON resolution envelope.
Verification level: A verification level is AgenID's label for how an agent's identity was checked, from a registered self-declaration (L1_REGISTERED) up through domain, organization and deployment checks. Levels above L1 are expressed as signed verification assertions.

Cryptography

Ed25519: Ed25519 is a public-key signature scheme, specified in RFC 8032, that uses the edwards25519 curve with 32-byte public keys and 64-byte signatures. AgenID signs every proof and assertion with pure Ed25519 over RFC 8785 canonical JSON.
Manifest digest: A manifest digest is the SHA-256 hash of an agent manifest's RFC 8785 canonical bytes, recorded as manifest_digest inside AgenID's signed objects. It binds each signature to exactly one version of the manifest.
ManifestProof: A ManifestProof is the operator-signed object that binds an AgenID identifier to one version of its manifest by SHA-256 digest. It is signed with an operator-role Ed25519 key and can only ever produce the DECLARED claim state.
RFC 8785 (JSON Canonicalization Scheme): RFC 8785, the JSON Canonicalization Scheme (JCS), defines one exact byte serialization for any JSON value, with sorted members, fixed number formatting and no insignificant whitespace. AgenID canonicalizes every hashed or signed object with it.

Keys and discovery

Authority key: An authority key is an Ed25519 key with role authority, held by a verification authority and used to sign verification assertions. In AgenID no authority key has been created yet, so no assertion can be issued.
Domain Connect: Domain Connect is an open standard that lets a service provider ask a user's DNS provider to apply a predefined set of DNS records, described in a template, after the user signs in and consents.
Key rotation and revocation: Key rotation replaces a signing key with a new one and retires the old key; key revocation declares a key no longer trustworthy from a point in time. In AgenID, retired and revoked key documents stay resolvable forever.
Operator key: An operator key is the Ed25519 key pair an agent operator uses to sign ManifestProofs. Its public half is published as a key document with role operator and a controller naming the agent; its private half never leaves the operator.
Root authority key: The root authority key is the Ed25519 trust anchor at the top of AgenID's verification model: the key whose signature makes an assertion above L1_REGISTERED valid, and whose public value verifiers pin out of band. It has not been created yet.
Two-path key discovery: Two-path key discovery is AgenID's method for obtaining an operator's public key from two independent sources, the AgenID registry and the operator's own domain, and comparing them. Agreement means a single compromised source cannot swap a key unnoticed.
Well-known URI: A well-known URI is a URI whose path begins with /.well-known/, a prefix defined by RFC 8615 so that clients can find metadata about an origin at a predictable location before making other requests.

Agents and ecosystems

Agent Card (A2A): In the Agent2Agent (A2A) protocol, an Agent Card is a JSON metadata document an agent publishes to describe its identity, capabilities, skills, service endpoint and authentication requirements so other agents can discover it.
Agentic commerce: Agentic commerce is buying and selling in which an AI agent, rather than a person clicking through a checkout, searches, compares, negotiates or pays on someone's behalf.
Model Context Protocol (MCP): The Model Context Protocol is an open-source standard for connecting AI applications to external systems such as data sources, tools and workflows.
Web Bot Auth: Web Bot Auth is an IETF working group, and the family of drafts it develops, for cryptographically authenticating automated clients such as crawlers and agents when they access websites.

Policy and regulation

EU AI Act Article 50: Article 50 of the EU AI Act sets transparency obligations, including that providers design AI systems meant to interact directly with people so those people are informed they are interacting with an AI system.
Know Your Agent (KYA): Know Your Agent (KYA) is an emerging industry term for the checks a business runs before letting an AI agent act on its systems: which agent it is, who operates it and what it may do. It borrows its shape from Know Your Customer programs in finance.