What is a non-human identity?

Why NHIs are harder to govern than people

Each of these breaks an assumption that human identity governance depends on.

• No natural lifecycle — nothing equivalent to an employee leaving triggers deprovisioning.
• No owner by default — the creator moves on and accountability evaporates.
• Credentials that do not expire, copied into config, CI and code along the way.
• Permissions granted for the hardest case the software might ever face, then never reduced.
• Invisible to access reviews built around named human reviewers.

What agents change

A traditional NHI is predictable: a batch job that reads the same table nightly. Its behaviour is a narrow, stable distribution, which makes anomalies easy to spot and least privilege easy to define.

An agent is not predictable by design. Its purpose is to decide what to do. It may call a tool it has never called, reach a record it has never touched, or take an action in response to content it just read. Least privilege cannot be derived from observed behaviour alone, and behavioural baselining gets much weaker.

Agents also act on behalf of someone. A batch job represents only itself; an agent inherits context from a person or another agent, and that delegation chain is part of its effective authority.

Where NHI programmes and agent security meet

NHI tooling is strong at discovery, ownership and credential hygiene — finding keys, flagging stale accounts, rotating secrets. That work is necessary and applies directly to agents.

It generally stops at the grant. It answers what an identity is permitted to do, not whether a specific action should proceed now, given who asked, what content the agent just processed, and whether the action is reversible. Agent security is largely about that second decision, which is why the two disciplines are converging rather than competing.

Common questions

What counts as a non-human identity?

Service accounts, API keys, OAuth clients, workload identities, certificates and tokens — any identity used by software rather than a person.

Why do non-human identities outnumber human ones?

Every integration, pipeline, job and service tends to create at least one, and few are ever removed, because nothing equivalent to an employee leaving triggers deprovisioning.

Is an AI agent just another non-human identity?

It is one, but an unusual one. A traditional NHI has predictable behaviour, so least privilege can be derived from observation. An agent decides what to do, and acts on behalf of a person or another agent, so its effective authority includes a delegation chain.