Pular para o conteúdo principal

Configure network content filtering with file policies (Preview)

Implementation Effort: Medium

User Impact: Medium

Overview​

In a Zero Trust security model, organizations must protect sensitive data at every layer—including the network edge—to prevent unauthorized data exfiltration regardless of user location or destination. Network content filtering with file policies extends data loss prevention capabilities to the network layer through Global Secure Access, enabling real-time inspection and control of file transfers to prevent data leaks to unsanctioned destinations such as generative AI applications, unmanaged cloud apps, and untrusted internet sites.

This solution combines Microsoft Purview's advanced data classification service with Global Secure Access's identity-centric network security policies to create a comprehensive network-layer Data Loss Prevention (DLP) solution. By integrating content inspection with real-time user risk evaluation and identity-aware policies delivered through Conditional Access, organizations can enforce granular controls over sensitive data movement without compromising productivity or security posture.

Implementation steps:

  • Enable Internet Access traffic forwarding profile and assign users/groups
  • Configure TLS inspection policy (prerequisite for inspecting encrypted traffic)
  • Verify Global Secure Access client is properly routing internet traffic
  • Create file policies with basic actions (Allow/Block) or "Scan with Purview" for advanced DLP
  • Configure matching conditions (activities: upload/download, file types: MIME types, destinations: FQDNs/categories)
  • Link file policies to security profiles with appropriate priority
  • Create Conditional Access policies targeting "All internet resources with Global Secure Access" and assign security profiles
  • Test policy enforcement by attempting file uploads/downloads to configured destinations
  • Monitor enforcement through Traffic logs and DLP alerts

Important considerations:

  • Microsoft Purview license required for "Scan with Purview" action (basic file policy works without Purview)
  • Apps may use multiple URLs/FQDNs—ensure all relevant destinations are configured for policies to take effect

Restrict AI File and Text Transfers with Basic Network Content Policies​

Implementation Effort: Medium - Select transfer types/destinations, attach policies, and test application compatibility.
User Impact: Medium - Broad file/text rules can break AI applications, APIs, and legitimate uploads/downloads.
Documentation checked: 2026-11-11
Ownership: Shared network baseline checkpoint; reuse equivalent existing controls.

Overview​

Global Secure Access (GSA) basic content policies allow or block supported file MIME types and text types for selected destinations. They can restrict unneeded transfers to AI services without requiring Purview inspection.

These controls are different from sensitive-data classification, malware scanning, and prompt-injection detection. Blocking an executable MIME type does not establish that all transferred files are safe.

This supports Use least privilege access by limiting unnecessary transfer types and Assume breach by reducing permitted data-transfer channels.

Prerequisites​

  • Complete AI_163, including Internet Access acquisition and TLS inspection.
  • Internet Access licensing and the administrative roles needed for content policies/profile assignment.
  • Application owners have approved required and prohibited transfer scenarios.
  • Basic content filtering does not require a Purview license; Scan with Purview does.

Implementation guidance​

  1. Inventory file and text transfers used by the AI service or supported agent path.
  2. Under Global Secure Access > Secure > Content policies, create destination-scoped rules with Allow or Block actions for supported content types.
  3. Prefer narrow rules and pilot destinations. Broad JSON/HTML text blocking can interrupt ordinary web/API traffic.
  4. If using web categories as destinations, satisfy the documented web-filtering-policy prerequisite.
  5. Attach the policy through the supported profile path. For Copilot Studio, verify the policy is supported and apply AI_027.
  6. Test prohibited and approved transfer types, request/response behavior, and application-specific upload methods.
  7. Review exceptions and changes whenever an AI application changes its transfer format.

Validation and completion evidence​

  • A prohibited synthetic transfer is blocked; an approved transfer succeeds.
  • GSA logs identify the responsible content policy and action.
  • JSON/HTML/API workflows required for normal operation remain functional.
  • Unsupported transfer encodings and paths are documented with compensating controls.

Scope and limitations​

File detection supports unencoded request/response bodies and multipart encoding. Encoding files inside JSON or splitting them across encrypted requests can prevent inspection.

WebSocket traffic bypasses file-type filtering, although text content type is still evaluated. Do not infer file coverage from successful text filtering.

Basic Allow/Block content policies use GSA traffic logs. Purview alerts and Activity explorer apply only when Scan with Purview is used; basic MIME/text blocking does not itself generate Purview DLP evidence.

Reference​