मुख्य सामग्री पर जाएं

Configure AI Destination, Source-Type, and HTTP-Method Filtering

Implementation Effort: Medium - Define AI destination rules, supported source conditions, policy assignments, and pilot tests.
User Impact: Medium - Restricted destinations or methods can disrupt legitimate AI applications and agent workflows.
Documentation checked: 2026-11-11

Overview​

Global Secure Access (GSA) web content filtering controls access by web category, URL, and FQDN. AI governance can combine approved destinations with restrictions on unapproved generative AI services.

Web content filtering also supports source traffic type conditions for Agent versus User, and HTTP-method conditions. These support differentiated agent policies and limits on selected operations. They do not replace application authorization: POST can carry read or write operations, and a network-visible method does not fully describe tool intent.

This supports Use least privilege access by limiting reachable services and operations, and Verify explicitly through policy decisions on acquired traffic.

Prerequisites​

  • Complete AI_163 readiness and verify acquisition for the target source.
  • Confirm GSA administrative roles, Internet Access licensing, current feature availability, and supported clients.
  • Review approved AI services and exceptions with business and data owners.

Implementation guidance​

  1. Under Global Secure Access > Secure > Web content filtering policies, Switch to V2 profiles, create policies for approved/restricted AI categories, URLs, and FQDNs.
  2. Use specific destination exceptions where an entire category is too broad. For FQDN coverage, include the root domain as well as *.domain.example where both are needed.
  3. Where supported, set Source type to Agent or User for differentiated rules. Client-based classification requires task/process metadata; remote networks do not support this condition.
  4. Where needed, restrict HTTP methods such as PUT, PATCH, or DELETE for selected destinations. Enable TLS inspection for HTTPS method enforcement.
  5. Attach the policies through the feature's supported security-profile path. For Copilot Studio use AI_027; do not apply client-only source classification assumptions to that integration.
  6. Review rule priority, approved exceptions, and the DNS/browser/transport prerequisites in the linked configuration guide before rollout.

Validation and completion evidence​

  • An approved destination succeeds and a prohibited destination is blocked.
  • If source conditions are used, test the same destination from agent and user contexts.
  • If method conditions are used, test allowed and blocked operations against a controlled endpoint, not production data.
  • Logs identify the expected policy decision; applications remain functional for approved use.

Scope and limitations​

Without TLS inspection, HTTPS filtering is limited to SNI-based rules rather than payload-level method/source inspection described in the current prerequisites. Source traffic type requires a supported client path. URL filtering has a documented tenant limit of 1,000 URLs.

These controls do not inspect sensitive prompt content; use network DLP. They do not restrict individual MCP tools; use MCP firewall.

Reference​