Activate Baseline Security Policies for Copilot Studio Agent Traffic
Implementation Effort: Low - Attach approved policies and verify runtime decisions after routing is established.
User Impact: Medium - Tenant-wide baseline changes can affect supported agent and other Internet Access traffic.
Documentation checked: 2026-11-11
Overview
Copilot Studio's Global Secure Access (GSA) integration currently supports enforcement through the baseline security profile. Security profiles linked to Conditional Access policies are not supported for this integration.
Creating filtering policies or enabling forwarding alone does not prove enforcement. Attach and enable the applicable policies in the baseline profile, then verify the exact agent interactions they are intended to govern.
This supports Verify explicitly by turning configuration into observed enforcement and Assume breach by removing unfiltered paths within the supported integration scope.
Prerequisites
- AI_022 is enabled, and AI_023 is complete for relevant existing custom connectors.
- Applicable web, threat-intelligence, and content policies are reviewed and ready.
- A nonproduction test environment and a recovery procedure are available.
Implementation guidance
- Under Global Secure Access > Secure > Security profiles, open Baseline profile and select Edit.
- Use Link a policy > Existing policy to attach each policy supported for the intended Copilot Studio traffic.
- Confirm linked policies are enabled. Review policy/rule ordering and the impact on other tenant traffic.
- Save the profile, allow for propagation, and run tests through actual agent actions.
- Verify an approved destination/action and a prohibited destination/action, recording matching logs.
- Include baseline attachment and runtime verification in the same rollout as policy creation. It is not a later optional optimization.
Validation and completion evidence
- The baseline profile lists the intended enabled policies.
- An in-scope agent request is blocked by the expected policy, and approved agent behavior remains functional.
- Logs correlate the agent schema name, destination, policy, and observed outcome.
- Tenant-wide impact and rollback ownership are recorded.