メインコンテンツへスキップ

Govern Generative AI Apps by Risk and Enforce Access Decisions

Implementation Effort: Medium - Review application risk, establish enforcement integrations, and validate device coverage.
User Impact: Medium - Blocked apps require sanctioned alternatives and a documented exception process.
Documentation checked: 2026-11-11

Overview​

Microsoft Defender for Cloud Apps provides cloud-app discovery, risk information, and Sanctioned/Unsanctioned classifications. These decisions become effective access controls only when connected to an enforcement mechanism.

An Unsanctioned tag alone is not universal network blocking. With the Defender for Endpoint integration enabled and configured, domains associated with unsanctioned apps are propagated to endpoint controls. Supported security-appliance integrations or exported block lists provide alternative enforcement paths.

This supports Verify explicitly by tying risk decisions to approved AI use and Assume breach by reducing exposure to unapproved services. It complements GSA category and destination filtering rather than replacing it.

Prerequisites​

  • Existing AI app discovery and an agreed risk/approval process.
  • Appropriate Defender for Cloud Apps and enforcement-product licensing.
  • For the Defender for Endpoint path: onboarded supported devices, required cloud protection/network protection settings, administrative access, and the documented integration prerequisites.
  • An inventory of unmanaged, unsupported, or otherwise unenforced devices and traffic paths.

Implementation guidance​

  1. Review discovered AI apps using business need, data handling, compliance, and risk information. Approve alternatives before blocking widely used tools.
  2. Select the enforcement path and document its actual device/network scope.
  3. For Defender for Endpoint, enable Enforce app access under Defender portal Settings > Cloud Apps > Cloud Discovery > Microsoft Defender for Endpoint.
  4. Enable Custom network indicators under Settings > Endpoints > Advanced features and verify required endpoint protection settings.
  5. Mark prohibited apps Unsanctioned manually or through an approved discovery policy. Scope enforcement and warning experiences where supported.
  6. Allow up to three hours for an Unsanctioned app decision to propagate to endpoint blocking, then test enforcement. Do not assume immediate enforcement.
  7. Align decisions with AI_025 GSA filtering. Resolve conflicting allow/block policies and establish a review/exception process.

Validation and completion evidence​

  • A prohibited AI app is blocked on an in-scope device; an approved alternative remains accessible.
  • The app decision, enforcement configuration, endpoint coverage, and observed block are recorded.
  • Unmanaged or unsupported paths have an owner and compensating control.
  • Blocking is verified independently of the discovery dashboard or Unsanctioned tag.

Scope and limitations​

Domain-based enforcement is not a guarantee of application-layer inspection, prompt DLP, or coverage for every agent host. Browser, OS, endpoint protection, and integration support must be checked against the selected enforcement path.

Reference​