Shadow AI Detection Lands in Microsoft 365 Admin Center, Enforced Through Intune

Microsoft's new Shadow AI page detects unmanaged AI agents like OpenClaw and lets admins block them via an auto-deployed Intune policy.

AI & AUTOMATION

8/8/20262 min read

a man riding a skateboard down the side of a ramp
a man riding a skateboard down the side of a ramp

Microsoft has rolled out a dedicated Shadow AI page in the Microsoft 365 admin center, giving IT admins a way to actually see which unmanaged AI agents are running on their fleet instead of guessing. The feature, part of the Frontier preview program, sits under Agents > Shadow AI and is separate from the existing All agents view — it's built specifically to surface consumer-facing AI tools that employees installed on their own, with no IT visibility or approval. Microsoft frames the risk in familiar terms: data leakage, compliance gaps, security exposure, and a general lack of auditability once these agents are running unsupervised on a managed endpoint.

During public preview, the detectable agent list covers OpenClaw, ChatGPT Desktop, Ollama Desktop, Poe Desktop, Claw/ZeroClaw, OpenCode, and Claude Desktop. Detection works across all of them, but blocking is currently limited to OpenClaw — and only on managed Windows devices enrolled in Microsoft Intune. Microsoft says the agent list and behavior are still subject to change before the feature reaches general availability, so admins shouldn't treat today's coverage as final.

Getting access requires stacking a few prerequisites: a Microsoft 365 E5 license to view Shadow AI agents, at least one qualifying Entra role (Security Administrator, AI Administrator, Global Reader, Security Reader, Security Operator, Reports Reader, User Experience Success Manager, or Intune Administrator), Intune enrollment on the target Windows devices, and an opt-in to the Frontier preview experience in the admin center. Admins who also enable Global Secure Access get a fuller picture — GSA unlocks the traffic tile (unique endpoints, requests, data sent/received) plus the users, most recent activity, and last scanned fields in the agent detail pane. Without GSA, those fields stay empty.

The enforcement mechanism is where this ties directly into existing Intune workflows. Blocking an agent doesn't happen through a separate console — selecting Security policies > Block > Apply Policies on a detected agent like OpenClaw generates a standard Intune policy (named "A365 - Block OpenClaw") that propagates to all enrolled Windows devices the normal way. Microsoft's own documentation notes that, depending on tenant configuration, that propagation can take anywhere from 15 minutes to 8 hours. Because it lands as a regular Intune policy, admins can find it, inspect it, and layer additional controls onto it using the same device configuration tooling they already use for everything else — no bolt-on management surface to learn.

For admins who want to get ahead of this before it's fully baked, the practical first step is confirming Frontier program eligibility and E5 licensing, then opting in and checking which of the seven detectable agents already show activity in the tenant. Given that blocking currently only works for OpenClaw, the near-term value is mostly visibility — knowing that ChatGPT Desktop or Ollama is running on a dozen laptops is useful intelligence even before a one-click block exists for it. Expect the blockable agent list to grow as the preview matures, and it's worth revisiting the Shadow AI page periodically rather than treating this as a one-time setup task. As with most Frontier-tier previews, the feature set, detection accuracy, and rollout pace are Microsoft's to change without much notice, so admins piloting it should treat findings as directional rather than authoritative until it reaches general availability.

Connect

Stay updated with latest Microsoft blog news

© 2025. All rights reserved.