Non-Human Identities: Why Identity Alone Does Not Establish Intent

Non-human identities#

Non-human identities are used by software, including AI agents, service accounts, API clients, workloads, bots, integrations, and automation processes. They may have a token, certificate, key, role, or delegated authority. Those controls answer who or what presented the request. They do not fully answer whether the request is appropriate in its current context.

That gap becomes more important when an AI agent can browse, call tools, use a customer's session, retrieve data, or take a transaction action. A valid credential can be stolen, over-scoped, reused by an unexpected process, or exercised outside the task that the organization intended to authorize.

Intent#

Identity establishes a starting point. Security teams also need to establish the authority, requested action, scope, and conditions that apply to that identity. A support agent that reads an order has different authority from one that changes a recovery email. A catalog agent and an agent that submits a payment should not receive the same policy.

Define the permitted actions, data access, transaction limits, time window, user approval requirement, and escalation path for each high-risk task. Then verify those conditions as the request develops. The AI agent verification guide explains this as continuous verification of identity, intent, and context.

Behavior#

Behavioral context supplies evidence that credentials cannot. Request timing, tool use, navigation, device and network changes, session state, transaction value, and action sequence can show whether an agent is operating within its expected journey. A sudden API sweep, recovery request, destination change, or high-value transaction may need a new decision even when the initial identity was recognized.

hCaptcha Bot Detection can detect and identify many AI agents and evaluates behavioral, device, network, and intent signals in real time. Its Rules Engine can apply an organization's business logic to identified agent traffic or all automation. That lets a team set separate conditions for browsing, login, account recovery, or a transaction.

Continuous trust#

Trust should change when the agent's circumstances change. An organization can allow a low-risk action, request verification for an uncertain one, limit repeated activity, hold a sensitive request, or deny an action that violates policy. The response should match the evidence and the consequence of the action.

hCaptcha User Journeys uses a blinded user ID to connect behavioral, device, and network signals across signup, login, authenticated sessions, APIs, and transactions. It can help analysts investigate a recognized agent as it approaches a sensitive action while the organization retains the mapping to its own customer identity.

The companion guide to AI agent security risks covers prompt injection, delegated authority, data exposure, and other conditions that can change the risk of an agent action after its identity has been established.

Implementation#

Inventory every non-human identity and the systems it can reach. Document the owner, credentials, authority, permitted tools, data classes, sensitive actions, expiration, monitoring, and revocation process. Start with the identities that can access customer data, change account controls, move money, or administer infrastructure.

Test representative journeys: an approved request, an expired or over-scoped credential, an unexpected tool call, an unusual session change, a limit breach, and a recovery or incident path. Keep the policy decision and supporting evidence with the event so security, fraud, and support teams can review an outcome and adjust the control.

Frequently asked questions#

What is a non-human identity?

A non-human identity is an identity used by software, such as an AI agent, API client, service account, workload, bot, or automation process. It commonly uses a key, token, certificate, or delegated permission.

Why is non-human identity security important for AI agents?

AI agents can act across applications, tools, sessions, and customer workflows. A valid identity needs boundaries around its authority, data access, actions, and conditions so the organization can limit misuse or unexpected behavior.

Does identity verification prove an AI agent is safe?

No. It verifies the identity presented to the service. The organization still needs to assess the action, authority, session context, behavioral evidence, and risk of the outcome.

How can teams verify an AI agent continuously?

Reassess the agent when it reaches a sensitive action or its context changes. Use policy rules, behavioral and session signals, action limits, verification, logging, and escalation procedures that fit the potential harm.

How does hCaptcha help with non-human identities?

hCaptcha can identify many AI agents and evaluate automation with behavior, device, network, intent, session, and journey context. Teams can use the resulting evidence in their own rules to allow, verify, limit, hold, or block an action.