Copilot AI Governance

Checks GET /identityProtection/riskyAgents (graph-rest-beta — confirmed via the Microsoft.Graph.Beta.Identity.SignIns PowerShell module reference as the documented endpoint for this resource; no v1.0 equivalent exists today). The riskyAgent resource carries riskState (none, confirmedSafe, dismissed, atRisk, confirmedCompromised, unknownFutureValue), agentDisplayName, riskLevel, riskDetail, isEnabled, and isDeleted.

"At risk and not yet remediated or dismissed" maps specifically to riskState eq 'atRisk': confirmedSafe and dismissed are both post-investigation resolved states. confirmedCompromised agents are listed separately in the result (never silently dropped) but don't drive this finding's title, since they represent an already-confirmed worse state rather than an unremediated pending one.

Uses the least-privileged permission confirmed in the Microsoft Graph permissions reference, IdentityRiskyAgent.Read.All — a distinct permission from IdentityRiskyServicePrincipal.Read.All, which covers a different, non-agent resource. The Learn page for this feature currently states Identity Protection for agents will require a Microsoft Agent 365 license "starting soon" (not required at the time this test was written).

A 403 is reported as NotAuthorized with a message calling out the IdentityRiskyAgent.Read.All permission and the possibility that this tenant has already started requiring a Microsoft Agent 365 license. A 404 is reported separately (also skipped, not a pass) with a message noting the beta endpoint may not yet be rolled out in this tenant, or Identity Protection for agents may not be enabled — distinct from an unrelated failure, which still skips with the generic Error path.

[!NOTE] Validated 2026-11-06 against a real Microsoft 365 test tenant (F7-licensed), after the user pre-consented the IdentityRiskyAgent.Read.All scope. The run correctly Passed (true negative): GET /identityProtection/riskyAgents returned 0 records in this tenant, so neither the atRisk nor confirmedCompromised paths were exercised against real data this round.

Remediation action

Investigate each agent identity flagged atRisk and either remediate the underlying risk signal or dismiss it once reviewed. Investigate confirmedCompromised agents immediately — they represent a risk investigation that has already concluded against the agent.

Related links

• Get-MgBetaRiskyAgent
• Microsoft Graph permissions reference

%TestResult%