Who should own review and offboarding for service accounts and AI agents?

Ownership should sit with the team that relies on the identity to run production work, with identity governance defining the policy and evidence requirements. If ownership is left ambiguous, offboarding slows down, rotation stalls, and stale credentials stay active long after their original purpose has ended.

Why This Matters for Security Teams

Ownership is the control point that determines whether a service account or AI agent is treated as a managed production identity or as a leftover credential after the team moves on. Identity governance can define standards, but the operating team that depends on the identity is the only group with enough context to judge whether the workload is still live, whether the secret is still needed, and whether the blast radius has changed. That distinction matters because stale identities are a recurring cause of exposure, as highlighted in The 2025 State of NHIs and Secrets in Cybersecurity from Entro Security.

For autonomous agents, the issue is sharper. An agent may spin up tools, request new scopes, and chain actions far beyond a static service account’s original purpose, which is why current guidance increasingly aligns with runtime governance in OWASP Agentic AI Top 10 and the NIST AI Risk Management Framework. In practice, many security teams encounter offboarding failures only after a retired workload is still authenticating in production, rather than through intentional identity retirement.

How It Works in Practice

The cleanest model is shared ownership with clear boundaries. The product or platform team that runs the workload owns day-to-day review, business justification, and offboarding execution. Identity or security governance owns policy, minimum evidence, approval thresholds, and exception handling. That split keeps accountability close to the system while preserving consistent control across the estate.

For service accounts, review should answer four questions: is the workload still active, is the account still scoped correctly, is rotation happening on schedule, and can the account be retired or replaced with a stronger identity pattern. For AI agents, the same review must also consider the agent’s tool access, delegation chain, and whether runtime authorization should replace static standing privileges. NHIMG’s NHI Lifecycle Management Guide and Top 10 NHI Issues both emphasize that lifecycle ownership is not a paperwork task, it is an operational control.

• Assign a named technical owner for every NHI, service account, and agent identity.
• Require business purpose, system dependency, and fallback owner before approval.
• Make review cadences based on risk, privilege, and whether the identity can act autonomously.
• Use time-bound credentials and revoke access automatically when the workload ends.
• Route offboarding through change, incident, and asset retirement processes so retired identities are not missed.

Implementation should also align with runtime policy checks and workload identity primitives, especially where agents use ephemeral tokens, SPIFFE-based trust, or other short-lived credentials. These controls tend to break down when ownership is split across multiple platform teams without a single accountable production operator, because no one has end-to-end visibility into when the identity should be retired.

Common Variations and Edge Cases

Tighter ownership controls often increase review overhead, so organisations must balance stronger accountability against the speed required by production systems and AI-driven pipelines. That tradeoff is real, especially when teams use shared platforms, ephemeral environments, or delegated agent frameworks that create identities faster than humans can manually review them.

There is no universal standard for exactly which team must sign off in every case, but best practice is evolving toward “own the workload, govern the policy.” Security should not be the operational owner unless it also runs the service, because central teams usually lack the context to know whether an identity is safe to retire. Likewise, platform teams should not be allowed to self-approve indefinite access without evidence of active use and clear offboarding triggers. The CSA MAESTRO agentic AI threat modeling framework reinforces that agent ownership must include lifecycle and misuse scenarios, not just login control.

Edge cases include shared service accounts, vendor-managed agents, and break-glass identities. These need explicit exception owners, expiration dates, and revalidation dates. For high-risk environments, pairing this model with the NIST AI Risk Management Framework and NHIMG’s Ultimate Guide to NHIs helps formalize who reviews, who approves, and who proves offboarding actually happened.

Related resources from NHI Mgmt Group

• Who should own governance when access spans humans, service accounts, and AI agents?
• How should organizations approach the governance of AI agents?
• What problem does ownership attribution solve for service accounts and API keys?
• When do service accounts become a higher risk than ordinary user accounts?