Use Compliance Frameworks to determine Duo Agent Platform availability in projects

Problem to solve

Organizations managing regulated projects or projects handling sensitive data currently lack a scalable mechanism to control GitLab Duo Agent Platform access based on compliance requirements. While group-based access controls exist, they don't leverage the existing compliance framework infrastructure that organizations already use to classify and govern their projects.

Current limitations:

  • Duo Agent Platform access is controlled through manual group assignment or instance/namespace toggles
  • No automatic relationship between a project's compliance posture and AI tool access
  • Organizations must maintain separate governance structures for compliance frameworks and AI access
  • No way to enforce "AI-free zones" for projects under specific regulatory requirements (e.g., HIPAA, PCI-DSS, SOC2, customer contractual obligations)

Example scenarios requiring this capability:

  • A financial services company needs to disable Duo Agent Platform for all projects labeled with their "PCI-DSS" compliance framework due to contractual restrictions on AI tool usage
  • A healthcare organization wants to enable Duo Agent Platform only for internal tooling projects while restricting it from patient data systems marked with their "HIPAA" framework
  • An enterprise needs to phase Duo Agent Platform adoption by first enabling it only for projects without compliance framework requirements

Intended users

Proposal

Extend compliance frameworks with the ability to control GitLab Duo Agent Platform availability at the project level.

Configuration Location: Add a new "AI Tool Access" section when creating or editing compliance frameworks at:

  • Group level: Secure > Compliance center > Frameworks
  • Centralized compliance frameworks: Compliance and security policy group

New Settings:

AI Tool Access (Optional) ○ Default (inherit from group/instance settings) ○ Enabled (allow Duo Agent Platform regardless of higher-level settings) ○ Disabled (block Duo Agent Platform for projects with this framework)

Enforcement Logic:

  1. When a compliance framework is applied to a project, the framework's AI tool access setting takes precedence over group/namespace settings
  2. A "Disabled" framework blocks Duo Agent Platform even if enabled at the group/instance level
  3. An "Enabled" framework allows Duo Agent Platform if the user has proper permissions and it's enabled at instance level
  4. "Default" inherits existing group-based and namespace settings
  5. Multiple frameworks on a project: Most restrictive setting wins (Disabled > Default > Enabled)

Benefits:

  • ✅ Leverages existing compliance framework infrastructure
  • ✅ Provides audit trail through compliance center reporting
  • ✅ Scales automatically as new projects are added to frameworks
  • ✅ Aligns AI governance with existing compliance workflows
  • ✅ Visible in compliance dashboards showing AI access status

User experience goal

As a Compliance Manager, I want to control AI tool access based on regulatory requirements so that I can ensure projects under specific compliance obligations don't use AI assistance when contractually restricted, while still allowing AI adoption for appropriate projects.

Example workflow:

  1. Create compliance framework "PCI-DSS Level 1" with AI Tool Access = "Disabled"
  2. Apply framework to 50 payment processing projects
  3. All 50 projects automatically block Duo Agent Platform access
  4. Compliance dashboard shows: "50 projects with AI access restricted by compliance framework"
  5. Developers see clear message: "Duo Agent Platform is disabled for this project due to PCI-DSS Level 1 compliance requirements"

Permissions and Security

  • Framework AI settings management: Requires Owner role at the group level (consistent with current compliance framework permissions)
  • No permission escalation: Framework settings cannot grant access beyond what instance/group admins have configured
  • Audit logging: Framework-based restrictions logged in audit events

Documentation

Would require updates to:

Availability & Testing

Tier: Ultimate (aligns with compliance framework feature tier)

Feature flag: compliance_framework_duo_access_control (enabled by default)

Testing plan:

  • Unit tests for framework access logic
  • Integration tests for multi-framework inheritance
  • E2E tests for UI changes in compliance center
  • Access control verification in Duo Agent Platform features

Related issues

This complements existing access controls:

Edited by 🤖 GitLab Bot 🤖