Agentic Identity IndexAgent identity glossary
Read definitions of agent identity, delegation, credential brokering, runtime policy, human approval and MCP terms used in rankings and official documentation.
- Agentic identity
- The identity and access context of an AI agent acting on resources.
- Agent principal
- A distinct identity used to identify and authorise an agent.
- Agent owner
- The person or team accountable for an agent's access and lifecycle.
- Agent discovery
- Finding deployed or connected agents and their access paths.
- Shadow agent
- An agent operating outside the recorded inventory or approved workflow.
- Non-human identity
- An identity assigned to software, services or machines rather than a person.
- Workload identity
- An identity assigned to a running software workload.
- Delegation chain
- The sequence of identities and grants connecting a user to an agent action.
- On-behalf-of access
- A request made by an agent using the authority of a named user.
- Least privilege
- Granting only the access needed for a specific task.
- Short-lived credential
- A credential with a limited validity period.
- Standing secret
- A persistent credential that remains usable between tasks.
- Credential broker
- A service that supplies credentials after evaluating identity and policy.
- Secret rotation
- Replacing a credential and retiring its previous value.
- Token exchange
- Exchanging one security token for another with a defined audience or scope.
- Audience
- The intended recipient of a security token.
- Scope
- A permission or bounded set of permissions attached to an access grant.
- Runtime policy
- A rule evaluated while an agent requests access or executes an action.
- Per-action authorisation
- An access decision for each individual operation rather than a whole session.
- Human approval
- A human decision required before an action is allowed.
- Step-up authentication
- An additional identity check for a sensitive request.
- MCP
- Model Context Protocol, a protocol connecting clients to tools and resources.
- MCP gateway
- An intermediary that mediates agent access to MCP servers or tools.
- Tool allowlist
- A list of tools an agent is permitted to call.
- Tool call
- An agent request to execute a specific exposed operation.
- Consent
- A user's explicit grant of access for an application or agent.
- OAuth 2.1
- A consolidated OAuth authorisation framework used for delegated access.
- OIDC
- OpenID Connect, an identity layer built on OAuth 2.0.
- SPIFFE
- A framework for assigning verifiable identities to workloads.
- SPIRE
- An implementation of SPIFFE workload identity issuance and attestation.
- Cross App Access
- An OAuth extension for controlled app-to-app and agent-to-app access.
- Identity provider
- The system that authenticates identities and issues identity assertions.
- Audit trail
- Records that identify the actor, action, resource and policy outcome.
- Revocation
- Withdrawal of a credential, grant or permission before further use.
- Kill switch
- A control that disables an agent or removes its access.
- Access review
- A periodic check of who or what should retain permissions.
- Attestation
- Evidence used to establish the identity or state of a workload.
- Policy decision point
- The service that decides whether a request is allowed.
- Policy enforcement point
- The component that applies the access decision to a request.
- Credential isolation
- Keeping downstream credentials outside the agent's own execution context.