Inactive agent identities can be disabled and eventually deleted under Microsoft’s public-preview Entra Agent ID lifecycle policies. The feature, highlighted in Microsoft’s 6 November Entra roundup, combines a no-licence default inactivity policy with separately licensed custom rules for reconfirming an identity’s purpose, maintaining sponsors and monitoring inactivity.
The issue is access that outlasts the work it was created for. Microsoft’s lifecycle-policy documentation describes a governance process that evaluates identities, notifies accountable people and enforces configured actions when a violation remains unresolved. Microsoft says the preview is provided without a service-level agreement and is not recommended for production workloads.
Entra Agent ID defaults act at 60, 90 and 455 inactive days
The default-policy documentation says no licence is required and the policy is available to all tenants. Once enabled, it periodically evaluates the agent identities in its scope. At 60 days of inactivity it notifies sponsors and owners; at 90 days it disables the identity; at 455 days it deletes it. These are elapsed inactivity thresholds, not three additional waiting periods to add together.
The default actions and timeframes are not customisable. Administrators can, however, select all Agent IDs or exclude selected identities, and configure fallback notification recipients. Microsoft says excluded identities can remain inactive indefinitely, accommodating seasonal or limited-use agents. Availability to all tenants therefore is not the same as enforcement over every AI agent in every application.
The docs also distinguish sponsorship from recovery authority. If the default policy has disabled or deleted an identity, only an owner or administrator can recover it; sponsors cannot re-enable or restore it. Accountability before an action and authority to reverse it are different permissions.
That distinction sits alongside IBM’s agent identity and preview kill-switch controls. The Microsoft policy concerns lifecycle thresholds and enforcement for scoped identities, rather than an assertion that every agent process can be stopped immediately across all services.
Custom rules can test continued need, sponsorship and inactivity
For custom policies, periodic reconfirmation can be set between 30 and 730 days. Sponsors or owners confirm that an identity is still required. A separate rule specifies the minimum number of sponsors, while inactivity monitoring checks whether no tokens have been issued for the identity during the configured duration. This is identity-token activity, not a measure of how many words a language model generates.
Available actions are disable, delete through soft deletion, or disable followed by deletion after a configurable grace period. Unremediated violations trigger the selected process. Where custom policies overlap, the highest-priority matching policy applies; active custom policies take precedence over the default inactivity policy.
Microsoft lists the licence options as Microsoft 365 E7, which includes Agent 365 and Microsoft Entra Suite, or a Microsoft Agent 365 licence paired with at least Microsoft Entra P1 or Microsoft 365 E3. That requirement belongs to custom governance capabilities, not the no-licence default policy described above.
The broader ownership problem also appears in Mintive’s per-agent identity proposition and Workday’s agreement to acquire C1.ai for identity access management. These address related enterprise access questions without establishing that their agents are automatically covered by Microsoft’s policy.
An identity lifecycle rule is different from a prompt instruction
The important boundary is where the decision is enforced. Asking an agent in chat to stop using an account leaves the request inside the agent’s conversational workflow. A lifecycle policy instead evaluates an identity against configured rules and invokes a governance action. The model is not being asked to decide whether its own identity should continue to exist. That makes the mechanism relevant to stale access even when the original task conversation has ended.
It is not a substitute for deciding which identities belong in scope. An exclusion can preserve a legitimate seasonal agent, but it also changes which identities receive the default cleanup. Sponsor requirements and reconfirmation ask different questions from recent activity: an active identity can still need an accountable person and a continuing business purpose. Treating those as separate rules avoids equating activity with permission to continue indefinitely.
The preview warning matters at that same operational level. A configurable enforcement mechanism is consequential precisely because disabling an identity can interrupt its use. Microsoft’s lack of an SLA and recommendation against production workloads remain material limits on how the announcement should be read, rather than a guarantee of production-ready coverage.
The next practical milestone is validation on a limited test scope, which Microsoft’s documentation describes before broader enforcement.
Sources
- Microsoft Entra roundup, 6 November 2026
- Microsoft Learn: Entra Agent ID lifecycle policies
- Microsoft Learn: default inactivity policy
Related stories
IBM Makes Agent Identity GA; Its AI Agent Kill Switch Is in Preview compares another identity-control approach while keeping its generally available and preview components separate.
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
