What breaks when access is reviewed only on a schedule for NHIs and agents?

Why schedule-based reviews miss the real access decision

Scheduled recertification assumes access is a stable property of an identity. For NHIs and agents, that assumption breaks down because authority is often created, consumed, and removed inside the same workflow. A monthly or quarterly review can still satisfy a governance calendar while completely missing the moment when a token, scope, or delegated permission actually enabled action.

That gap matters because the control is judging an entitlement snapshot, not runtime authority. If the identity only existed for minutes, or if its permissions were narrowed and expanded by automation during execution, the review may confirm yesterday’s state while the system has already moved on.

This is why scheduled review works better for slow-changing human access than for high-churn machine access. The control can still be useful as a backstop for ownership, exception handling, and inventory hygiene, but it is a weak primary control when access is transient or task-scoped.

What actually breaks in governance and assurance

The first break is timing. When access is granted just in time, exchanged through a token flow, or inherited through a delegated workflow, the meaningful question is whether the actor was allowed to do something at the instant of use. A later certification cannot prove that, so the review may preserve a false sense of closure.

The second break is scope drift. NHIs and agents often accumulate broad standing permissions to avoid operational friction, then only use a narrow slice of them. Scheduled review may see an approved role or entitlement and miss that runtime behaviour is narrower, broader, or different from the documented intent.

The third break is accountability. If no one owns the live decision path, reviews become administrative evidence instead of operational control. That is why ownership, rotation, and offboarding remain important, but they do not substitute for observing authorization where it is actually consumed.

Why the problem gets worse with agents and ephemeral workloads

Agents and other non-human actors often chain several actions inside one task, and each action may require different scopes, tools, or downstream service access. That means the access pattern is dynamic by design. A schedule-based review can still confirm that a permission exists, but it cannot tell you whether the permission was appropriate for each sub-action or whether the agent exceeded its intended task boundary.

Ephemeral credentials create a similar blind spot. Short-lived tokens, federated credentials, and workload identities are often designed to disappear before the next review cycle. If governance only checks on a calendar, the most security-significant state may no longer exist by the time anyone looks.

For that reason, good practice is to treat scheduled review as an assurance layer, not the place where authorization is decided. The control plane needs runtime signals, task boundaries, and revocation capability if you want the answer to reflect actual use rather than historical assignment.

Practitioner Guidance

What to verify: Verify whether the review process examines current entitlement state, or only a periodic snapshot. If the access can be created and consumed inside a single job, require an operational control that sees the live grant, not just the next certification cycle.

Decision rule: If the identity uses task-scoped or short-lived access, rely on runtime authorization, logging, and revocation triggers for primary control, and use scheduled reviews only as secondary governance evidence. If access is long-lived and static, scheduled review has more value, but it still should not be the only control.

Practitioner takeaway: The key failure is not that reviews are absent, it is that they are late. For NHIs and agents, governance must follow the moment of authorization use, or the control will certify state that no longer exists.

Related resources from NHI Mgmt Group

• What is the difference between reviewing human access and reviewing NHIs?
• What breaks when AI agents are granted SaaS access for testing and never reviewed again?
• What breaks when access reviews are the main control for AI agents and NHIs?
• How should security teams run access reviews for non-human identities?