Security teams should give each agent its own cryptographically verifiable identity, scope permissions to the workflow and task, and make delegation explicit at every hop. The control goal is traceability, revocation, and least privilege. Shared service accounts and borrowed user sessions hide accountability and widen blast radius, especially when agents can act at machine speed across APIs and databases.
What identity should an autonomous agent have in an enterprise workflow?
An autonomous agent should not inherit a human’s identity by default. It needs its own identity boundary so actions can be attributed to the agent, not to a shared account or a borrowed user session. That boundary also makes it possible to enforce policy consistently across approvals, tools, and downstream systems, instead of treating the agent as an invisible extension of a person.
For implementation detail, the identity model should be tied to how the agent is registered, authenticated, and retired. NHIMG’s Agentic AI Identity Guide is useful where teams need a practical view of identity models, delegation and lifecycle rather than a purely theoretical discussion.
How should delegation work across hops and tools?
Delegation should be explicit at every hop, with each step carrying only the authority needed for that specific task. The practical test is whether a reviewer can answer who granted access, what was delegated, to which tool or system, and for how long. If the answer is unclear, the workflow is relying on implicit trust instead of controlled delegation.
This is especially important when agents interact with APIs, databases, or orchestration layers that can turn one request into several downstream actions. NHIMG’s OWASP ASI03 Guide explains the same control objective through identity and privilege abuse, while RFC 8693: OAuth 2.0 Token Exchange is the standard reference for exchanging delegated authority rather than passing around a long-lived bearer token.
What operating model keeps agents safe at enterprise scale?
The safest operating model is to treat agent identity and delegation as part of workflow design, not as a late security overlay. That means choosing short-lived credentials, binding permissions to task scope, and making revocation easy when a workflow changes or an agent is retired. It also means separating human approvals from machine execution so the control plane can prove when authority was granted and when it ended.
NHIMG’s Top 10 Agent Identity Controls is a good companion for the control set, and Agent Identity and Access Guide helps teams map those controls to identity models, delegation choices and lifecycle governance.
Risk and Threat Considerations
When autonomous agents run with borrowed identities or broad standing access, the main risks are attribution failure, privilege sprawl and blast-radius expansion. A compromise can look like legitimate automation unless the delegation chain is explicit, which makes detection and rollback slower.
Failure mechanism: Shared credentials, token passthrough without audience limits, or opaque delegation hops let an agent act with authority that outlives the original business decision. That can create confused-deputy behaviour, unsafe lateral movement across systems, and weak auditability when multiple actions are chained together.
Impact: Teams may lose the ability to prove who authorised an action, revoke access cleanly, or contain misuse to a single workflow. At scale, this increases the chance that one agent failure becomes an enterprise-wide access incident.
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.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | Agent identity and explicit delegation are central to this workflow control. |
| Recommendation — Scope each agent to task-bound authority and verify every delegated hop before execution. | ||
| NIST SP 800-53 Rev 5 | IA-9 — Service Identification and Authentication | Autonomous agents authenticating to systems need distinct machine-to-machine identity controls. |
| AC-6 — Least Privilege | Task-scoped permissions and blast-radius reduction are the core access objective here. | |
| AU-2 — Event Logging | Traceability and revocation depend on auditable records of delegated actions. | |
| Recommendation — Authenticate each agent as its own service identity and avoid shared credentials. Limit each agent to the minimum permissions needed for the current workflow step. Log agent actions, delegation events and revocation points for later review. | ||
Practitioner Guidance
What to verify: Confirm that every production agent has a unique identity, a documented owner, and an auditable delegation path. If an agent can still operate after the human initiator’s session ends, you need a separate control for its authority, not a borrowed login.
Decision rule: If a task can change records, move funds, or query sensitive data, require task-scoped delegation and a revocation path before go-live. If the workflow cannot support that discipline, reduce the agent’s authority or keep the action human-approved.
Practitioner takeaway: The objective is not to make agents “trusted”, but to make every meaningful action attributable, bounded and revocable even when the workflow is fully automated.
Related resources from NHI Mgmt Group
- How should security teams implement session persistence and identity enforcement for AI agents in stateful enterprise workflows?
- How should security teams govern AI agents that can access enterprise systems?
- How should security teams implement zero trust for autonomous agents in financial workflows?
- How should security teams implement identity controls for autonomous AI agents across APIs and human-facing interfaces?
Deepen Your Knowledge
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.
Reviewed and updated by the NHIMG editorial team on November 5, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org