Using Defender for Cloud Apps to control access to Generative AI

Question

We have Microsoft Defender and enabled the Cloud Apps component to discover. One of our goals is to see what Generative AI tools our users are using and start to use Sanction / Unsanctioned. For some AI tools, we have Sanctioned for all our employees. However, we have some users (mainly Developers) that will need to use additional AI tools but not the rest of employees. What are the steps to controlling access so that a specific group of users is allowed to use specific AI tools but is blocked for the rest of our users?

Thx in advance for any help given.

Answer 1

Use Defender for Cloud Apps for org‑wide discovery and blocking, then add more granular controls with Microsoft Entra Internet Access and (optionally) Intune.

A practical approach in three layers:

1. Discover and classify AI apps
   1. In the Microsoft Defender portal, go to Cloud apps → Cloud app catalog.
   2. In Category, select Generative AI.
   3. Review risk scores and usage, and tag apps as Sanctioned or Unsanctioned according to organizational policy.
      • See the guidance in Step 1: Discover AI apps for using the catalog and risk scores.
2. Block unsanctioned AI apps for the whole organization
   1. In Defender for Cloud Apps, after identifying AI apps that should be blocked for most users, select them in the Cloud app catalog.
   2. Mark them as Unsanctioned. This:
      • Blocks them at the organizational level via Defender for Cloud Apps network controls.
      • For devices onboarded to Defender for Endpoint, unsanctioned apps are automatically blocked (unless the security team configures “warn and educate” instead of hard block).
   3. Optionally, create an app governance policy that targets unsanctioned generative AI apps:
      • Policy template: No template
      • Apps matching all of the following: Category equals Generative AI AND Tag equals Unsanctioned
      • Apply to: All continuous reports This gives a default posture of “blocked for everyone” for those AI tools.
3. Allow specific AI apps only for specific users/groups Org‑wide unsanctioning is coarse‑grained. To allow exceptions (for example, developers) while keeping everyone else blocked, use Microsoft Entra Internet Access and related controls:
   1. Keep the AI apps Unsanctioned in Defender for Cloud Apps so they are blocked by default.
   2. Use Microsoft Entra Internet Access Conditional Access policies to selectively allow access:
      • Create policies that:
        • Target specific users or groups (for example, a “Developers” security group).
        • Allow access to the specific AI app domains for those groups.
        • Keep the default block in place for all other users.
   3. For elevated‑risk users, combine with Microsoft Purview Insider Risk Management Adaptive Protection:
      • Users flagged as high risk can be automatically restricted from accessing AI apps, even if they belong to a normally allowed group.
4. (Optional) Enforce device‑level controls To prevent users from bypassing browser‑based controls by installing native AI clients:
   • Use Microsoft Intune to block installation of unsanctioned AI apps on managed devices, following the guidance in “Blocking and removing apps on Intune managed devices”.

Resulting model:

• Defender for Cloud Apps: discovers all generative AI apps, provides risk scores, and enforces org‑wide “sanctioned/unsanctioned” posture.
• Microsoft Entra Internet Access: narrows access so only specific users/groups (for example, developers) can reach selected AI tools.
• Intune (optional): blocks installation of unsanctioned AI apps on devices.

---

References:

• Manage generative AI apps for your organization
• Step 1: Discover AI apps
• Step 2: Block access to unsanctioned AI apps
• Protect your organization’s use of AI apps and data
• Discover AI apps and data