A governance relationship that records whether an AI agent acts independently, on behalf of a human, or through another agent. It is important because accountability, approval, and risk ownership can change depending on who or what the agent represents at runtime.
Expanded Definition
An agent-on-behalf-of relationship specifies who the agent is representing at runtime and under what authority it is operating. In NHI governance, this is not just a label for delegation. It is an auditable control relationship that helps determine whether an AI agent may take actions for itself, act for a human, or delegate through another agent in a chain of execution.
The distinction matters because approval boundaries, logging requirements, and liability can change based on the represented party. Definitions vary across vendors, especially where multi-agent orchestration and delegated workflow execution are involved, so organisations should treat this as a governance assertion rather than a purely technical property. In practice, the relationship should be explicit, machine-readable, and tied to identity context, consent, and scope. That is consistent with the governance emphasis in the NIST AI Risk Management Framework and the action-oriented guidance in OWASP Agentic AI Top 10.
The most common misapplication is assuming an agent inherits a human's authority automatically, which occurs when delegated intent is not bound to a specific scope, duration, or approval state.
Examples and Use Cases
Implementing agent-on-behalf-of rigorously often introduces workflow friction, requiring organisations to weigh automation speed against clearer approval and accountability boundaries.
- A support agent drafts responses on behalf of a service desk analyst, but only within the analyst's approved queue and subject to human review before send.
- A code-generation agent acts on behalf of a developer to open pull requests, while repository write permissions remain constrained by scoped delegation and audit logs.
- An orchestration agent invokes another agent to complete a task, but the parent relationship remains visible so investigators can trace which identity made the final decision.
- A procurement workflow allows an agent to prepare purchase requests on behalf of a business user, yet finance approval still requires a separate human signer.
- An identity platform records when a machine agent is operating on behalf of a human versus independently, supporting investigations and policy enforcement informed by the Ultimate Guide to NHIs and the AI LLM hijack breach.
These patterns are also shaped by guidance in the CSA MAESTRO agentic AI threat modeling framework and the MITRE ATLAS adversarial AI threat matrix, especially where nested delegation can obscure intent.
Why It Matters in NHI Security
Agent-on-behalf-of relationships are central to determining ownership when an agent creates, modifies, approves, or exfiltrates data using delegated access. If the relationship is missing or ambiguous, security teams cannot reliably answer who authorised the action, which policy applied, or whether the agent exceeded its mandate. That uncertainty becomes operational risk during incident response, access reviews, and legal review.
The scale of the NHI problem makes this especially important: NHI Mgmt Group reports that 97% of NHIs carry excessive privileges, which means delegated agents often inherit more access than intended. When the delegated actor is also an agent, that privilege becomes harder to reason about and easier to misuse. Strong relationship tracking supports least privilege, traceability, and zero trust decisioning, especially when paired with the OWASP Top 10 for Agentic Applications 2026 and the NIST AI Risk Management Framework.
Organisations typically encounter the consequences only after a delegated agent makes an irreversible change, at which point the relationship record becomes operationally unavoidable to address.
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 and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST AI RMF set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | N/A | Agent delegation and authority boundaries are core concerns in agentic AI security guidance. |
| OWASP Non-Human Identity Top 10 | NHI-01 | Relationship tracking supports governance over non-human identities and their access paths. |
| NIST AI RMF | The framework requires mapped accountability, oversight, and risk management for AI actions. |
Document delegated authority, approvals, and accountability for every agent-on-behalf-of relationship.
Related resources from NHI Mgmt Group
Deepen Your Knowledge
Reviewed and updated by the NHIMG editorial team on June 20, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org
