Control 3.8: Copilot Hub and Governance Dashboard
Control ID: 3.8 Pillar: Reporting Regulatory Reference: FINRA 4511, SEC 17a-3/4, GLBA 501(b), SOX 404 Last UI Verified: February 2026 Governance Levels: Baseline / Recommended / Regulated Last Verified: 2026-02-03
Objective
Establish centralized governance of Microsoft 365 Copilot and Copilot Studio agents through the unified Copilot Hub spanning M365 Admin Center and Power Platform Admin Center. This control provides visibility into AI adoption, configuration management, and agent lifecycle governance.
Feature Status (February 2026)
The Copilot Hub governance capabilities are evolving. Some features have reached GA while others remain in preview. Feature availability, naming, and functionality may change.
Feature Status Tracking:
| Feature | Status (Feb 2026) | Expected GA | Workaround for GA-Only |
|---|---|---|---|
| Copilot Hub (M365 Admin) | GA | GA | N/A |
| Copilot Hub (PPAC) | GA | GA | N/A |
| Agent Registry | Preview | TBA | Manual inventory tracking |
| Entra Agent ID | Frontier Preview | TBA | Service principals / managed identities |
| Conditional Access for Agents | Preview (depends on Entra Agent ID) | TBA | Standard CA policies |
| M365 Admin Center Agent Settings | GA | GA | N/A |
| Agent 365 Unified Control Plane | Preview (Frontier) | TBA | Per-platform admin centers |
| Agent 365 Observability | Preview (Frontier) | TBA | Platform-specific analytics |
| MCP Server Governance | GA (Oct 2025) | GA | N/A |
Agent Registry Visibility: Copilot Studio agents are visible. Microsoft Foundry agents expected at GA. Declarative agents appear but lack org-wide deployment — export/import required for distribution.
Check Microsoft 365 Roadmap and Message Center for GA announcements.
Terminology Clarification
- Copilot Hub (Admin) - The governance interface in M365 Admin Center and PPAC for administrators to manage Copilot settings, agents, and policies. This is the focus of this control.
- copilot.cloud.microsoft - The end-user portal (formerly Microsoft Copilot) where users access M365 Copilot chat. This is NOT a governance interface.
Why This Matters for FSI
- FINRA 4511: Usage records support books and records requirements
- SEC 17a-3/4: Agent activity documentation must be preserved
- GLBA 501(b): Access control for customer data requires centralized oversight
- SOX 404: IT controls for AI systems require documented governance
- Examination Readiness: Copilot configuration evidence for regulators
Control Description
The Copilot Hub provides unified governance across M365 Admin Center (Copilot and Agents sections) and PPAC (Copilot section). It enables centralized settings management, agent inventory, approval workflows, and usage monitoring.
| Portal | Section | Primary Functions |
|---|---|---|
| M365 Admin | Copilot | Settings, connectors, billing |
| M365 Admin | Agents | Registry, approvals, MCP Servers |
| PPAC | Copilot | Copilot Studio settings, usage |
| Dynamics 365 Sales Admin | Copilot | Sales-specific Copilot settings |
| Dynamics 365 Admin | Copilot | Dynamics 365 Copilot configuration |
| Power Pages Admin | Copilot | Power Pages Copilot settings |
Per-Product Copilot Hub Pages
In addition to the primary M365 Admin Center and PPAC hubs, Microsoft provides per-product Copilot Hub pages for Dynamics 365 Sales, Dynamics 365, and Power Pages. Each product hub surfaces product-specific Copilot settings and usage analytics. FSI organizations should include these product-level hubs in governance reviews when the corresponding products are deployed.
Key Capabilities:
| Capability | Description |
|---|---|
| Copilot Settings | Organization-wide AI configuration |
| Agent Registry | Complete inventory of all agents |
| Approval Workflow | Control agent publication |
| MCP Server Governance | Manage agent capabilities |
| Usage Reporting | Track Copilot adoption and usage |
January 2026 Enhancements (MC1187780)
New Admin Center Capabilities
The following capabilities were announced in MC1187780 (January 2026) and are rolling out to tenants.
Security Pivot on Copilot Overview Page:
A new Security pivot has been added to the Copilot overview page in M365 Admin Center, providing quick access to security-related Copilot settings and status.
Readiness Page:
A new Readiness page organizes Copilot configuration settings into three logical categories:
| Category | Settings Included |
|---|---|
| Deployment Essentials | License assignment, user enablement, rollout planning |
| End-User Experience | Web search, plugins, agent access, personalization |
| Data Security | DLP policies, sensitivity labels, audit configuration |
Success Metrics:
The Readiness page displays key adoption metrics:
| Metric | Description |
|---|---|
| Chat Active Users | Users actively engaging with Copilot Chat |
| Assisted Hours | Time saved through Copilot assistance |
| Satisfaction Rate | User satisfaction scores (if feedback enabled) |
These enhancements consolidate governance visibility that was previously spread across multiple admin center locations.
Copilot Hub Product Pages (GA)
The Copilot Hub has been restructured into product-specific pages providing targeted AI usage and governance insights:
- Copilot Studio page: Agent creation, usage, and cost metrics for Copilot Studio agents
- Power Apps page: AI features usage within Power Apps
- Power Automate page: AI-assisted flow creation and AI Builder usage
- Settings page: Centralized Copilot Hub configuration
- What's New page: Feature announcements and update notifications
Navigate to admin.powerplatform.microsoft.com > Analytics > Copilot Hub to access product pages.
ROI Analytics and Usage Insights (GA)
- Time-Cost Savings Tracking: Copilot Hub now provides ROI analytics (GA since October 2025) showing estimated time savings, cost avoidance, and productivity gains from Copilot and agent usage across the tenant. These metrics support executive reporting and business justification for AI investments
- Analytics Themes: Administrators can categorize Copilot usage patterns using Analytics Themes, grouping interactions by business function, department, or use case to identify high-value adoption patterns and governance gaps
AI Feature Access Control
GA Feature
AI Feature Access Control settings in the M365 Admin Center Copilot Hub are generally available. These settings provide granular user-level and feature-level controls for Microsoft 365 Copilot.
The Copilot Hub provides admin controls for managing user-level access to M365 Copilot features. These capabilities enable FSI organizations to restrict feature availability based on license assignment, role, and compliance requirements.
| Feature | Status | Description | Configuration |
|---|---|---|---|
| License-Based Restrictions | GA | Control Copilot access via M365 license assignment | M365 Admin > Licenses > Assign/Remove |
| Admin Exclusion Groups | GA | Exclude specific users/groups from Copilot features | M365 Admin > Copilot > Settings > User access |
| Copilot Chat Pinning | GA | Control whether Copilot Chat is pinned in Teams/Outlook | M365 Admin > Copilot > Settings > End-User Experience |
| Deployment Groups | GA | Staged rollout of Copilot to specific user groups | M365 Admin > Copilot > Settings > Deployment |
| Web Search Control | GA | Enable/disable Copilot web grounding per organization | M365 Admin > Copilot > Settings > Data access |
| Agent Access Control | GA | Control which agents users can discover and use | M365 Admin > Copilot > Settings > Actions |
Zone-Based AI Feature Enablement
Organizations implementing AI governance across the three-zone model should align AI feature access controls to zone risk profiles and regulatory requirements:
| Setting | Zone 1 (Personal) | Zone 2 (Team) | Zone 3 (Enterprise) | Regulatory Rationale |
|---|---|---|---|---|
| User Access | All licensed users | All licensed users | Specific user groups only | SOX 404: controlled access to IT systems |
| Admin Exclusion Groups | Not required | Compliance-sensitive roles | Traders, restricted persons, investigated employees | FINRA 3110: supervisory restrictions on communications tools |
| Deployment Groups | Optional phased rollout | Recommended phased rollout | Mandatory phased rollout with approval gate | SEC: controlled change management for customer-facing systems |
| Web Search | Enabled | Disabled for MNPI teams | Disabled organization-wide | GLBA 501(b): prevent external data leakage from customer interactions |
| Copilot Chat Pinning | User preference | Enabled for collaboration teams | Controlled per department | Adoption management aligned with supervision requirements |
| Agent Access Control | All agents allowed | Organizational + Microsoft verified | Organizational only, approved list | FINRA 4511: approved tools and systems only |
Recommended Rollout Path for FSI:
- Pilot Phase (Zone 1): Enable Copilot for IT and Compliance teams with all features enabled and comprehensive logging
- Wave 1 (Zone 1 expansion): Expand to non-customer-facing business units with Admin Exclusion Groups for compliance-sensitive roles
- Wave 2 (Zone 2 entry): Staged rollout to team collaboration environments with web search disabled and organizational agents only
- Wave 3 (Zone 3 production): Deploy to customer-facing environments with strictest controls, deployment groups gating access, and real-time supervision
Admin Exclusion Group Configuration
The CopilotForM365AdminExclude security group pattern enables administrators to exclude specific users from Copilot features in Microsoft admin centers (Microsoft 365 admin center, Exchange admin center, etc.) regardless of license assignment. This provides a compliance control for temporarily or permanently restricting admin-center AI tool access. Note: This exclusion group does not remove access to end-user Microsoft 365 Copilot features (e.g., Copilot in Word, Excel, Teams); those require separate license or policy controls.
Configuration Steps:
- Create a security group in Entra ID (Microsoft Entra admin center > Groups > New group)
- Name the group
CopilotForM365AdminExclude(exact name required) - Set Group type to "Security"
- Add members: users or nested groups to exclude from Copilot access
- Exclusion takes effect within 24 hours of group membership change
FSI Use Cases:
| Scenario | Population | Duration | Regulatory Driver |
|---|---|---|---|
| Trading blackout periods | Traders, investment bankers | Temporary (quarterly) | SEC Regulation FD, insider trading prevention |
| Compliance investigations | Employees under review | Temporary (investigation duration) | FINRA 3110: enhanced supervision |
| Restricted persons lists | Individuals with conflicts of interest | Permanent or semi-permanent | FINRA 2111: suitability, conflict management |
| Customer-facing trial period | Sales, advisory teams | Temporary (pilot phase) | Risk management during controlled rollout |
Operational Notes:
- Exclusion group membership changes propagate within 24 hours (not real-time)
- Users in exclusion group retain M365 Copilot license assignment but cannot access Copilot features
- Nested groups are supported (allows synchronization with existing compliance distribution lists)
- Audit log captures group membership changes for compliance evidence (Entra ID > Audit logs)
Deployment Group Rollout Strategy
Deployment Groups enable phased rollout of Microsoft 365 Copilot to specific user populations, supporting controlled adoption aligned with organizational readiness and regulatory compliance validation.
Phased Rollout Pattern:
| Phase | Population | Duration | Validation Criteria | Rollback Procedure |
|---|---|---|---|---|
| Pilot | IT, Compliance, AI Governance team (10-50 users) | 4-6 weeks | Feature functionality verified, no compliance violations, positive user feedback | Remove users from deployment group |
| Wave 1 | Non-customer-facing business units (100-500 users) | 8-12 weeks | Usage metrics healthy, DLP policies effective, no audit findings | Disable deployment group, return to pilot only |
| Wave 2 | Customer-facing with supervision (500-2000 users) | 12-16 weeks | Supervision workflows validated, regulatory reporting functional | Reduce deployment group to Wave 1 scope |
| Wave 3 | Full organization rollout | Ongoing | Continuous monitoring, quarterly compliance review | Revert to deployment group model if issues arise |
Zone-Aligned Rollout Recommendations:
- Zone 1 environments: Deployment groups optional; consider for large-scale rollouts to manage support load
- Zone 2 environments: Recommended to validate team collaboration patterns and data exposure before full enablement
- Zone 3 environments: Mandatory; align deployment group phases with change control approval gates and regulatory notification requirements
Rollback Procedure:
If issues arise during rollout (compliance violations, excessive hallucinations, user resistance):
- Remove affected users from deployment group immediately (takes effect within 8 hours)
- Document rollback reason and affected user count
- Conduct root cause analysis before re-enabling
- Update deployment criteria and validation steps
- Resume phased rollout with enhanced controls
FSI Governance Guidance
AI Feature Access Control capabilities help support FINRA 4511 (restricting AI tool access to approved populations and supervised personnel), GLBA 501(b) (controlling data exposure paths via web search controls and external data access restrictions), SOX 404 (documented IT controls over AI feature rollout and change management), and SEC 17a-3 (records of AI tool access, configuration changes, and user population changes for books and records requirements).
Recommended Control Configuration for Regulated Environments:
- Admin Exclusion Groups: Create and maintain exclusion groups for traders (during blackout periods), employees under compliance investigation, and restricted persons lists
- Deployment Groups: Implement mandatory phased rollout with compliance review gates at each wave transition
- Web Search Control: Disable organization-wide for Zone 3 environments processing customer data or MNPI; consider selective enablement for Zone 1/2 with DLP enforcement
- Agent Access Control: Restrict to organizational agents only for Zone 3; require approval workflow for all agents that access customer or financial data
- Copilot Chat Pinning: Align with supervision requirements — disable pinning for unsupervised roles to reduce inadvertent AI usage without oversight
Monthly Governance Review:
- Export Admin Exclusion Group membership and validate against current restricted persons list
- Review deployment group assignments against approved rollout plan
- Audit web search usage logs for unauthorized external data access attempts
- Verify agent access restrictions remain aligned with approved agent inventory
Key Configuration Points
- Configure Copilot Settings with FSI defaults (disable web search, block external AI)
- Configure user-level feature access via Copilot Settings
- Set up Admin Exclusion Groups for compliance-sensitive users
- Configure deployment groups for staged Copilot rollout
- Review Agent Registry weekly for ownerless agents and pending requests
- Configure allowed agent types to restrict to organizational and Microsoft-verified
- Manage MCP Servers to block unnecessary data access capabilities
- Set up billing policies by department for cost allocation
- Configure PPAC Copilot Studio settings (disable Computer Use, restrict channels)
- Export usage reports monthly for compliance documentation
- Configure transcript access controls with role-based restrictions (supervisor-only access, agent creator exclusion)
- Enable transcript logging with 7-year retention for FINRA/SEC compliance
- Implement separation of duties controls preventing agent creators from accessing their own transcripts
- Configure immutable audit trail for transcript access events with Purview Audit (Standard or Premium)
- Export transcripts to immutable storage (Azure Immutable Blob Storage or compliance recording system)
PPAC Copilot Studio AI Feature Controls
The following AI feature toggles in Power Platform Admin Center and Copilot Studio control agent capabilities that may introduce data protection risks if enabled without governance. Review and configure each setting per zone requirements.
| AI Feature | Portal Path | Recommended Default | Zone 1 | Zone 2 | Zone 3 | Risk if Uncontrolled |
|---|---|---|---|---|---|---|
| AI Prompts | PPAC > Environment > Settings > Product > Features > AI prompts | Off | Optional (with review) | Off unless approved | Off unless approved | Unintended data exposure through AI-generated suggestions |
| Generative Actions | Copilot Studio > Agent > Overview > Orchestration | Off | Optional (with review) | Off unless approved | Off unless approved | Unauthorized system actions, prompt injection, data exfiltration |
| File Analysis | Copilot Studio > Agent > Settings > Generative AI > File processing | Off | Optional (with review) | Off unless approved | Off unless approved | Sensitive file content processed without classification awareness |
| Model Knowledge | Copilot Studio > Agent > Settings > Generative AI > Use model knowledge | Off | Optional for low-risk | Off unless approved | Off | Unverified/hallucinated content in responses to sensitive queries |
| Semantic Search | Copilot Studio > Agent > Settings > Generative AI > Use semantic search | Off | Optional for low-risk | Off unless approved | Off unless approved | Indirect information leakage through intent-based knowledge retrieval |
| Generative AI Features | PPAC > Environment > Generative AI features | Review per-environment | Enable selectively | Restrict by default | Restrict by default | Uncontrolled AI feature availability across environments |
| Move Data Across Regions | PPAC > Environment > Generative AI features | Off | Off unless required | Off | Off | Data residency violations, regulatory non-compliance |
| Bing Search | PPAC > Environment > Generative AI features | Off | Optional | Off | Off | External data leakage from customer interactions |
| Microsoft 365 Services | PPAC > Environment > Generative AI features | Review per use case | Enable selectively | Review with compliance | Review with compliance | Uncontrolled data access across M365 service boundaries |
AI Feature Governance for Regulated Environments
For Zone 2 and Zone 3 environments, disable all AI feature toggles by default and enable only through a documented approval process that includes:
- Business justification for the specific feature
- Data classification review (what data the agent accesses)
- Risk assessment documenting mitigating controls
- Compliance officer sign-off
- Quarterly re-attestation of continued need
Conversational Transcript Access Governance
Copilot Studio conversational transcripts contain user inputs, agent responses, and organizational context that may include confidential data, customer information, or material non-public information (MNPI). Proper transcript access governance supports regulatory compliance with communications supervision, records retention, and separation of duties requirements.
Granular Access Controls
Role-Based Access Restrictions:
Organizations should implement granular access control for transcripts aligned with regulatory supervision requirements:
| Access Level | Permitted Roles | Typical Entra ID Groups | Regulatory Rationale |
|---|---|---|---|
| Full Transcript Access | Compliance Officers, Supervisors, Audit Teams | Copilot-Compliance-Supervisors, Internal-Audit-Team |
FINRA 3110: supervisory review of electronic communications |
| Agent Performance Data Only | Agent Developers, Product Managers | Copilot-Studio-Developers, AI-Product-Managers |
SOX 404: separation of duties - developers cannot access production communications |
| No Access | Agent Creators/Operators (same individual) | Self-service exclusion | FINRA 3110: prevent self-supervision of communications |
| Read-Only Audit Access | External Auditors (time-limited) | External-Audit-FY2026 |
SEC 17a-4: examiner access for regulatory review |
Configuration Path: PPAC > Environment > Settings > Product > Features > Copilot Studio Agents > Transcript Access
Recommended Practice: Use Entra ID Conditional Access to enforce MFA and compliant device requirements for transcript access roles.
Transcript Retention and Supervision
Regulatory Retention Requirements:
| Regulation | Retention Period | Scope | Implementation Requirement |
|---|---|---|---|
| FINRA 4511 | 7 years (3 years readily accessible) | All agent conversations involving customer interactions or recommendations | Export transcripts to immutable storage (e.g., Compliance Recording) |
| SEC 17a-3/4 | 7 years | Communications related to securities transactions or advisory services | Transcript exports with tamper-evident audit trail |
| CFTC 1.31 | 5 years | Commodity futures/derivatives communications | Transcript retention with regulatory retrieval capability |
| SOX 404 | Align with IT records policy (typically 7 years) | Agent configurations and system change records | Transcript metadata (who accessed, when, configuration state) |
Retention Configuration Guidance:
- Enable transcript logging: PPAC > Environment > Settings > Product > Features > Copilot Studio Agents > Enable Transcript Storage
- Configure retention period: Set minimum 7-year retention for FSI environments (align with FINRA/SEC requirements)
- Export to immutable storage: Use Power Automate flows to automatically export transcripts to Azure Immutable Blob Storage or approved compliance recording systems
- Tag customer-facing transcripts: Apply metadata tags to differentiate customer-facing vs. internal-only conversations for retention policy enforcement
Supervision Workflow Integration:
Organizations subject to FINRA 3110 supervision requirements should implement automated transcript review workflows:
- Daily Export: Export new transcripts from Copilot Studio to supervision platform
- Risk-Based Sampling: Route high-risk keywords (e.g., "trade", "recommendation", "guaranteed return") to compliance review queue
- Supervisor Assignment: Assign transcripts to designated supervisors (not agent creators) for review
- Exception Escalation: Flag policy violations for escalated review and remediation
- Quarterly Attestation: Supervisors attest to completion of transcript review for audit evidence
Separation of Duties
Agent Creator/Operator Restrictions:
To support FINRA 3110 supervision requirements and prevent self-review of communications, organizations should implement the following separation of duties controls:
| Control | Implementation | Enforcement Mechanism |
|---|---|---|
| Prevent creator self-access | Agent creators cannot access transcripts of agents they built | Entra ID Conditional Access policy denying transcript access role to Copilot-Studio-Publishers group |
| Supervisor assignment | All agent transcripts assigned to independent supervisor | Automated workflow assigns supervisor based on agent's business unit (not creator's) |
| Restricted operator access | Users operating agents daily cannot access historical transcripts without supervisor approval | RBAC policy requires Supervisor-Approval claim for transcript access |
| Compliance-only export | Bulk transcript exports restricted to Compliance team | PPAC environment security role limits "Export Data" permission to Compliance Officers |
Recommended Implementation:
- Create two separate Entra ID groups:
Copilot-Studio-Publishers(agent creators/developers) — NO transcript accessCopilot-Compliance-Supervisors(supervisors/compliance) — Full transcript access- Configure PPAC environment security roles to enforce separation
- Use Conditional Access to block
Copilot-Studio-Publishersfrom transcript URLs - Document separation in AI Governance Policy and verify quarterly
Immutable Transcript Audit Trail
Tamper-Evident Logging:
To support SEC 17a-4(f) WORM (Write Once, Read Many) requirements and audit integrity, organizations should implement immutable audit trail capabilities for transcript access and modifications:
| Audit Event | Required Metadata | Retention | Regulatory Justification |
|---|---|---|---|
| Transcript Accessed | User ID, timestamp, transcript ID, access duration | 7 years | SEC 17a-4: audit trail of record access |
| Transcript Exported | User ID, timestamp, export scope, destination | 7 years | FINRA 4511: evidence of records production |
| Transcript Deleted | User ID, timestamp, transcript ID, deletion reason, supervisor approval | Permanent | SOX 404: change management audit |
| Access Policy Changed | Admin ID, timestamp, old/new policy, approval ticket | Permanent | FINRA 3110: supervision policy changes |
Implementation Options:
- Purview Audit (Standard): Enable Purview Audit (Standard) and retain Copilot Studio audit events for 180 days (extend to 7 years with retention policy)
- Purview Audit (Premium): Capture granular transcript access events with 10-year retention and intelligent insights
- Export to SIEM: Forward Copilot Studio audit logs to security information and event management (SIEM) for correlation with other security events
- Immutable Storage: Configure Azure Immutable Blob Storage with legal hold or time-based retention for exported transcript archives
Verification Steps:
- Confirm Purview Audit logging captures "Copilot Studio transcript accessed" events
- Verify audit retention policy set to minimum 7 years for FSI compliance
- Test transcript export integrity with SHA-256 hash verification
- Validate access policy changes trigger workflow approval and audit log entry
- Conduct quarterly audit log review for unauthorized transcript access attempts
Caveat: Organizations should verify that their chosen audit and storage configurations meet their specific regulatory obligations. This control provides technical capabilities to support compliance but does not ensure regulatory compliance without proper implementation and ongoing validation.
DLP for Agent Publishing Connectors
- Block agent publishing connectors via DLP policies: Apply Data Loss Prevention policies at the environment level to block connectors such as "Copilot Studio for Microsoft Teams" and "M365 Copilot channel" in environments where agent publishing should be restricted. This prevents unauthorized deployment of AI agents through unapproved channels. See Control 1.5 for DLP policy configuration details
Zone-Specific Requirements
| Zone | Requirement | Rationale |
|---|---|---|
| Zone 1 (Personal) | Baseline settings; monthly review; transcript access: team leads + compliance | Low-risk personal use; limited supervision requirements |
| Zone 2 (Team) | Approval required for shared agents; weekly registry review; transcript access: designated supervisors (not agent creators) | Team data exposure; enhanced supervision for collaboration |
| Zone 3 (Enterprise) | Strictest settings; real-time monitoring; executive reporting; transcript access: compliance officers only with 7-year immutable retention | Customer-facing, highest risk; full FINRA/SEC supervision requirements |
Roles & Responsibilities
| Role | Responsibility |
|---|---|
| Entra Global Admin | Configure Copilot settings, manage billing |
| AI Administrator | Manage Copilot settings and feature access (delegated) |
| Power Platform Admin | Configure PPAC Copilot, Copilot Studio settings; Configure transcript access restrictions |
| AI Governance Lead | Review agent registry, approve publications |
| Compliance Officer | Validate settings meet regulatory requirements; Review and approve Admin Exclusion Group membership for regulatory compliance; Access and review conversational transcripts for supervision (FINRA 3110) |
| Supervisor | Review assigned agent transcripts for compliance violations; Cannot access transcripts for agents they created (separation of duties) |
Related Controls
| Control | Relationship |
|---|---|
| 1.1 - Restrict Agent Publishing | Enforced via Agent Settings |
| 1.2 - Agent Registry | Centralized in Agents section |
| 3.2 - Usage Analytics | Usage reports integration |
| 3.7 - Security Posture | Complementary governance |
| 2.22 - Inactivity Timeout Enforcement | Inactivity timeout compliance scan results provide data for consolidated governance visibility |
Automated Validation: Agent Access Governance Monitor
For automated detection of overly permissive agent access configurations across Power Platform environments, see the Agent Access Governance Monitor solution.
Capabilities:
- Zone-based agent access compliance validation (Zone 1/2/3 requirements)
- Daily scheduled drift detection with baseline comparison
- Teams adaptive card alerts with severity classification
- Dataverse-persisted validation history for audit trail
- SHA-256 integrity-hashed evidence export for examination support
Deployable Solution: agent-access-monitor provides PowerShell validation scripts, Power Automate flow definitions, and Dataverse schema for persistent governance state.
Implementation Playbooks
Step-by-Step Implementation
This control has detailed playbooks for implementation, automation, testing, and troubleshooting:
- Portal Walkthrough — Step-by-step portal configuration
- PowerShell Setup — Automation scripts
- Verification & Testing — Test cases and evidence collection
- Troubleshooting — Common issues and resolutions
Advanced Implementation: Configuration Hardening Baseline
This control is covered by the Configuration Hardening Baseline, which consolidates SSPM-detectable settings across all 7 mapped controls into a single reviewable checklist with automation classification and evidence export procedures.
Advanced Implementation: Unrestricted Agent Sharing Detector
For continuous detection of overly permissive agent sharing configurations at the per-agent level, see the Unrestricted Agent Sharing Detector. Complements the Agent Access Governance Monitor (environment-level) with per-agent sharing violation detection, automated remediation, and exception management.
Governance Script: Zone Access Validation
Test-ZoneAgentAccess.ps1 validates M365 Admin Center agent access settings against zone-based governance policies. Checks agent access policy alignment (Zone 1: all agents, Zone 2: Org + Microsoft verified, Zone 3: Org only), admin exclusion group membership, staged deployment group configuration, and web search controls — with drift detection and adaptive card alerting via adaptive-card-zone-access-alert.json (in FSI-AgentGov-Solutions).
Script Location: scripts/governance/Test-ZoneAgentAccess.ps1
Verification Criteria
Confirm control effectiveness by verifying:
- Copilot Settings accessible with all four tabs (User access, Data access, Actions, Other)
- Web search for M365 Copilot disabled for compliance-sensitive environments
- Admin Exclusion Groups correctly exclude designated users
- Deployment groups limit Copilot access to approved user populations
- Agent Registry shows all agents with accurate owner information
- Pending agent requests reviewed and actioned within SLA
- Ownerless agents identified and assigned within 14 days
- Monthly usage reports exported and archived
- AI Prompts toggle is disabled in PPAC for Zone 2/3 environments (PPAC > Environment > Settings > Features)
- Generative Actions are disabled for all agents without documented approval (Copilot Studio > Agent > Orchestration)
- File Analysis is disabled for agents without documented data classification review
- Model Knowledge is disabled for agents handling sensitive data
- Semantic Search is disabled for agents without approved and scoped knowledge bases
- Generative AI features, Move Data Across Regions, and Bing Search are reviewed and restricted per-environment
- Conversational transcript access is restricted to authorized personnel
- DLP policies block agent publishing connectors in restricted environments
- Transcript access roles configured with separation of duties (agent creators excluded)
- Transcript retention policy set to minimum 7 years for FINRA/SEC compliance
- Transcripts exported to immutable storage with tamper-evident audit trail
- Purview Audit logging captures all transcript access events with 7-year retention
- Supervisor assignment workflow prevents agent creators from reviewing their own transcripts
- Quarterly review of transcript access audit logs for unauthorized access attempts
Additional Resources
- Manage Microsoft 365 Copilot
- Manage Copilot Agents in Integrated Apps
- Copilot Usage Reports
- Power Platform Copilot Hub
- Copilot Studio Governance
Microsoft Audit Reporting Tools
For enhanced Copilot/AI reporting beyond native M365 Admin Center capabilities, see:
- Microsoft Audit Reporting Tools Playbook - AI-in-One Dashboard and PAX (Portable Audit eXporter) for enterprise-scale analytics
Microsoft Agent 365 Strategic Context (Preview)
Preview Notice
Microsoft Agent 365 was announced at Ignite 2025 as the strategic evolution of agentic AI capabilities. The following reflects preview guidance that may change.
Microsoft Agent 365 represents Microsoft's unified platform for agentic AI across the Microsoft ecosystem, announced at Ignite 2025. For FSI organizations, this signals:
| Consideration | Current State | Future Direction |
|---|---|---|
| Governance Surface | Copilot Hub (M365 Admin + PPAC) | Unified Agent 365 governance (roadmap) |
| Agent Types | Copilot Studio, Declarative, Extensions | Converging to Agent 365 SDK model |
| Identity Model | Service principals, user delegation | Agent ID (Entra workload identities) |
Recommended Approach: - Continue implementing Copilot Hub governance controls as documented - Monitor Agent 365 roadmap for GA timeline and migration guidance - Plan for potential governance consolidation in 2026-2027 - Note that admins can block or delete declarative agents from the registry but cannot deploy them org-wide; export/import workflows are required for broader distribution
Agent Essentials Governance (Preview)
Note: The following resources are preview documentation and may change.
Microsoft's Agent Essentials provides additional governance visualization and planning tools:
- Visual Governance Map - Decision flow for agent governance settings
-
MCP Governance - Model Context Protocol server management in Agent settings
-
Microsoft Learn: Agent Visual Governance Guide (Preview) - Visual map for governance decision points
- Microsoft Learn: Agent Deployment Checklist (Preview) - Comprehensive governance checklist
Data Refresh Timing
Settings Propagation
Configuration changes made in the Copilot Hub may take up to 8 hours to fully propagate across the tenant. Plan changes accordingly and verify settings after the propagation window.
Updated: February 2026 | Version: v1.3 | UI Verification Status: Current