KPATH

What is non-human identity for AI agents?

Non-human identity is the practice of giving software callers, including AI agents, a verifiable identity of their own rather than a shared key or a borrowed human login. For agents it goes further: the identity must be bound to the person or system the agent acts for, and something must enforce what that identity may do on each call.

Identity tells you who is calling; enforcement is what happens when the answer is not good enough.

Why agents made this urgent

Non-human identity is not new. Service accounts, API keys and machine certificates have been around as long as services have called each other. What changed is the caller.

A normal application does what you wrote. An agent decides what to do, picks a tool, and acts with real permissions. When such a thing holds a shared API key, or runs under a human’s login, the key or the login is now doing whatever the agent decides, and the record shows a person or a service account rather than the agent. Nobody can say afterwards which agent did what, on whose authority.

So the first job is to give each agent an identity of its own, with an owner and a risk tier behind it, and to stop it carrying anything else.

Workload identity

Workload identity is the mechanism for issuing that identity to a running process rather than configuring it by hand. The workload asks for its identity when it starts and receives a short-lived credential it can present to anything that trusts the issuer, instead of holding a long-lived secret someone pasted into a config file.

The properties that matter for agents are the ones that make the identity hard to steal and cheap to check: it is issued automatically, it expires quickly, and a relying party can verify it rather than take it on trust. Enterprises get those properties from several different mechanisms, inside a cluster and across one, and most estates run more than one at a time.

KPATH works with the identities you already issue. Every call through it carries the agent’s identity with an owner and a risk tier. High-risk agents can be required to present a workload identity that your identity provider verifies, so a name alone is not enough for the actions that matter.

Principal binding

An agent’s own identity answers only half of the question. Most consequential agent actions happen on behalf of someone: a customer asking an assistant to move money, an employee asking a harness to change a repo, a partner system arriving from outside.

Principal binding ties the agent’s identity to the person or system it is acting for on this request. The call then carries two facts: which agent is calling, and whose authority it carries. Policy can act on both. When an agent acts for a person, KPATH policy can insist that the person is verified, not merely named, before a high-risk action proceeds.

This is the fact that is missing from most estates today. Application logs show that a service account touched a system. They do not show that a particular customer’s request, through a particular agent, was the reason.

No credentials in the agent

Identity and credentials are different things, and the difference decides the blast radius.

An agent that carries the real credentials for the systems it uses is a walking key ring. If it is compromised, through prompt injection or a poisoned tool description, every door it could open is open.

Under KPATH, agents carry their own identity and nothing else. On an allowed call KPATH adds the real credential for the target system; the agent never sees it. Those credentials are encrypted under a key held in your own vault or key service. KPATH is a crypto user, not a custodian. A compromised agent holds nothing that opens a door anywhere else, and stopping the agent removes the only thing it had.

Why identity alone is not enforcement

A verified identity answers who is calling. Three more things have to happen before the call is governed.

Something has to decide whether this identity, acting for this principal, may take this action on this target right now. That decision belongs with your identity provider and the policy you write there.

The decision then has to be applied on the call: refuse it, hold it for a person, add the credential only if it passes. An identity provider is not on the path of the call and cannot do this. A directory of identities, however complete, is a register.

Last, the identity, the principal, the decision and the outcome have to be recorded in a form an auditor can verify. Otherwise the identity work produces a better-labelled log and nothing more.

Where KPATH sits

Your identity provider decides. KPATH enforces. KPATH works with the identities you already issue and applies the decision per action, before the payload reaches the target, for the agent and, where policy requires, for the person it acts for.

It governs the request envelope only: identity, target, action, size, delegation chain. Of the payload, it reads only the values you declare for a rule, such as a payment amount.

Start in monitor mode: agents keep their framework, model and prompts, each sends its calls through KPATH with its own identity, and the inventory shows which identities are actually in use, including the ones nobody registered. Nothing is refused until you choose to enforce.

FAQ

Common questions

What is non-human identity?

It is an identity issued to a piece of software rather than a person: a backend service, a scheduled job, an MCP client, or an AI agent. It replaces shared API keys and borrowed user logins with an identity that can be verified, owned, and revoked on its own.

What is workload identity?

Workload identity is the automatic issue of a short-lived identity to a running workload. The workload obtains it when it starts rather than from a secret pasted into a config file, and any relying party that trusts the issuer can verify it. KPATH can require a high-risk agent to present a workload identity that your identity provider verifies.

What is principal binding?

It is tying the agent's identity to the person or system it is acting for on this request. An agent acting for a customer carries the agent's identity and the customer's, and policy can insist the person is verified before a high-risk action proceeds.

Why is identity alone not enforcement?

Identity answers who is calling. It does not decide whether this caller may take this action on this system right now, and it does not act on the decision. That needs a point on the path of the call that can refuse it, hold it for approval, or stop the chain.

Published Updated

The first step

Start in monitor mode. See everything before you enforce anything.

Deploy in monitor mode: observe only, enforce no policy, rewrite no agents. Flip to enforce by repointing egress. See the agents already running in your estate before you decide what to enforce.

Book a monitor-mode pilot See the platform