Join our Newsletter — 33% off our NHI Course
Home› Glossary› Agentic AI & Autonomous Identity› On-Behalf-Of Agent
Agentic AI & Autonomous Identity

On-Behalf-Of Agent

← Back to Glossary
By NHI Mgmt Group Updated October 31, 2026 Domain: Agentic AI & Autonomous Identity

An AI agent that acts using a human user’s delegated permissions, usually through OAuth tokens or similar credentials. The user remains the nominal identity owner, but the agent can explore and use available access more aggressively than the person might, creating a lifecycle and audit challenge.

What On-Behalf-Of Agent Means in Practice

An on-behalf-of agent sits between a person and the systems that person can already reach. Its defining feature is delegated authority, not independent ownership of access. That makes it useful for automation and assistance, but it also changes how organisations think about consent, scope, and attribution.

The key question is not whether the agent can act, but how much authority it inherits and how narrowly that authority is bounded. In practice, the same delegated path can be safe for one task and risky for another, depending on token scope, audience restrictions, and how much the agent can amplify the user’s original permissions.

Delegated Authority and Token-Based Access

Most on-behalf-of patterns rely on OAuth token exchange or a similar delegation mechanism. The human user remains the nominal identity owner, while the agent receives a token or assertion that lets it operate within a bounded trust chain. That is why the model is often described as acting with borrowed authority rather than standing in as a separate user.

This distinction matters because delegation is not the same as impersonation. A well-designed flow preserves who initiated the action, who is acting, and which downstream resource is being requested. RFC 8693: OAuth 2.0 Token Exchange is the core reference for token exchange patterns that support this kind of chained authority.

NHIMG’s Delegation Chains and Attenuated Authority for AI Agents is useful here because it focuses on how delegation depth, actor claims, and attenuated capability tokens keep an agent from inheriting more than it should.

Why the Permission Model Becomes Harder to Govern

An on-behalf-of agent can explore available access more aggressively than the person would, even when it is technically staying within scope. That is the governance challenge: the person may have legitimate standing access, but the agent can chain calls, expand search depth, or trigger actions in ways that are difficult to predict from the original user intent.

This creates a lifecycle problem as much as an access problem. The organisation needs to know when the delegated right started, what the agent was allowed to do, when the token should expire, and how to revoke or constrain that delegation if the agent behaves unexpectedly. NHIMG’s Agentic AI Identity Guide covers the broader identity lifecycle for agents, including delegation, registration, and retirement.

It also explains why human and machine access increasingly overlap. Human vs Non-Human Identity helps frame the boundary where human ownership ends and delegated agent use begins.

Audit, Attribution, and Control Boundaries

Because the user remains the nominal owner, organisations often over-assume that standard user audit logs are enough. They are not. An on-behalf-of flow needs attribution that can show the initiating user, the acting agent, the token exchange event, and the downstream resource context, otherwise reviews quickly become ambiguous.

That is especially important where the same agent can act across multiple tools or services. Without clear attribution, investigators may only see that a valid user consented at some point, not that an agent later exercised that access in a broader or less obvious way. NHIMG’s Confused Deputy in AI Agents Guide is a strong companion reference because it explains how delegated authority can be abused when the actor, the consent, and the downstream request drift apart.

The control boundary should therefore be treated as a policy problem, not just an authentication problem. The practical issue is whether the agent is constrained to the intended task, the intended audience, and the intended resource set, rather than simply being able to present a valid token.

Where On-Behalf-Of Agents Show Up Most Often

These patterns appear wherever assistants can take actions inside enterprise systems, cloud services, or consumer workflows that already trust the user. They are especially visible in agent platforms, delegated business applications, and tool-enabled assistants where the user grants access once and the agent can then operate repeatedly until the delegation is removed or expires.

For that reason, the subject is not just about identity format or token mechanics. It is about how delegated authority is used, how much it can expand, and how confidently an organisation can reconstruct what happened after the fact. NHIMG’s Microsoft Entra Agent ID Guide is a useful implementation lens for organisations using Entra-based agent governance, access packages, and logging.

For a broader security framing, RFC 8693: OAuth 2.0 Token Exchange remains the best place to anchor the delegation model itself, while operational guidance should focus on scoping, expiry, and traceability rather than assuming that user consent alone is sufficient control.

Risk and Threat Considerations

On-behalf-of agents create a material exposure when delegated access is broader than the original user intent or harder to observe than normal interactive use. The risk is not just theft of credentials, but misuse of legitimate delegated authority, especially when the agent can chain access across tools or make decisions faster and more persistently than the person behind it.

Failure mechanism: The agent inherits valid user authority, then uses that authority in a way the user did not directly supervise, creating confused-deputy behaviour, overreach, or difficult-to-detect secondary actions.

Impact: The result can be data exposure, unintended business actions, privilege amplification through chained requests, and audit records that are too weak to explain who actually caused the action.

Standards & Framework Alignment

This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.

OWASP Agentic AI Top 10 addresses the attack and risk surface, while NIST SP 800-53 Rev 5 sets the governance and control requirements practitioners need to meet.

FrameworkControl / ReferenceRelevance
OWASP Agentic AI Top 10ASI03 — Identity & Privilege AbuseOn-behalf-of agents hinge on delegated authority and privilege boundaries.
Recommendation — Constrain agent authority so delegated actions remain tightly scoped and attributable.
NIST SP 800-53 Rev 5IA-5 — Authenticator ManagementToken-based delegation depends on secure lifecycle handling for authenticators and tokens.
AC-6 — Least PrivilegeDelegated agent access should be bounded to the minimum needed for the task.
AU-2 — Event LoggingOn-behalf-of actions require auditability across user, agent, and token-exchange events.
Recommendation — Manage token issuance, rotation, revocation, and expiry with strict lifecycle controls. Limit delegated permissions so the agent cannot exceed the intended user task. Log delegation, token exchange, and downstream actions with clear actor attribution.

Practitioner Guidance

Why practitioners should care: Treat on-behalf-of access as delegated authority with an expiry date, not as a permanent extension of the user account. The design question is whether the agent’s token can be limited tightly enough that the agent can only do the work that was explicitly intended.

Common misunderstanding: A valid user consent screen or OAuth grant does not automatically make the whole downstream behaviour safe. The real control point is the combination of scope, audience, token lifetime, and what the agent can do after the first API call.

Practitioner takeaway: If you cannot explain which actions the agent can take, on which resources, and for how long, the delegation model is too broad.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Reviewed and updated by the NHIMG editorial team on October 31, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org