Agent Identity

AI agent identity is the persistent or ephemeral identifier that represents an autonomous AI agent as an acting principal within a system, separate from the human user or service that initiated the agent.

Agentic IAM is identity and access management designed for agents that reason, choose tools, delegate work, and act dynamically during execution—going beyond static role assignments.

Agent identity is the identifier, credential, or token that represents an AI agent as a unique actor in a system, enabling access decisions, audit records, and policy enforcement to be attached to that specific agent.

Agentic identity refers to the full identity context of an AI agent at execution time, including who initiated the agent, what task it was assigned, what tools it is permitted to call, and what credentials have been derived for the current task.

Non-human identity (NHI) refers to identities used by machines, services, scripts, bots, and AI agents rather than by human users, requiring distinct lifecycle management, scoping, and credential practices.

Machine identity is the credential or certificate that uniquely identifies a non-human workload—such as a server, container, service, or AI agent—allowing it to authenticate to other systems.

Workload identity is the identity assigned to a running compute workload—such as a container, function, or agent runtime—used to authenticate the workload to cloud services and APIs without long-lived secrets.

An agent workload is the running compute unit that hosts an AI agent's execution—such as a container, Lambda function, or process—which may invoke multiple tools and sub-agents during a single task.

Agent-to-tool access describes the access relationship between an AI agent and the external tools, APIs, or data systems it invokes during execution.

An agent-to-agent delegation chain is the sequence of identity handoffs when an orchestrator agent delegates a sub-task to one or more sub-agents, each of which may further delegate, creating a chain of acting principals.

On-behalf-of access describes when an AI agent invokes a tool or system using access granted in the context of a human user's task, where the agent is acting as a delegate of that user.

Autonomous tool execution is when an AI agent selects and invokes tools without per-call human approval, making access control at the tool-call boundary the primary safety control.