Control 2.10: Insider Risk Detection for Copilot Usage — Verification & Testing
Test cases and evidence collection for validating insider risk detection for Copilot and agent activity.
Test Cases
Test 1: Insider Risk Policy Activation
- Objective: Confirm insider risk policies for Copilot are active and processing
- Steps:
- Navigate to Microsoft Purview > Insider Risk Management > Policies
- Verify Copilot-related policies show "Active" status
- Run Script 1 to confirm policy configuration
- Check that policy indicators include both Copilot-specific signals and AI usage indicators
- Expected Result: All configured insider risk policies are active with AI usage indicators enabled
- Evidence: Policy status screenshot and PowerShell output
Test 2: Risky Agents Policy Verification
- Objective: Confirm the Risky Agents policy is active and covers all deployed agents
- Steps:
- Navigate to Microsoft Purview > Insider Risk Management > Policies
- Locate the auto-deployed Risky Agents policy
- Verify the policy scope includes all deployed Copilot Studio and Azure AI Foundry agents
- Confirm alert routing is configured to reach both compliance and agent deployment owners
- Review the policy thresholds for FSI appropriateness
- Expected Result: Risky Agents policy is active, scoped correctly, and alert routing is configured
- Evidence: Policy configuration screenshot; alert routing configuration
Test 3: AI Usage Indicator Functionality
- Objective: Verify AI usage indicators are active and producing signals
- Steps:
- Navigate to Microsoft Purview > Insider Risk Management > Settings > Policy indicators
- Confirm AI usage indicator category is enabled
- Generate above-normal Copilot and agent activity volume with a test account
- Wait for the processing cycle (24-48 hours)
- Check the test user's risk timeline for AI usage indicator signals
- Expected Result: AI usage indicators appear in the risk timeline for elevated-activity accounts
- Evidence: IRM risk timeline showing AI usage indicator signals
Test 4: Anomaly Detection Functionality
- Objective: Verify the system detects anomalous Copilot usage patterns
- Steps:
- Generate above-normal Copilot activity volume with a test account
- Wait for the insider risk processing cycle (24-48 hours)
- Check for risk alerts or elevated risk scores on the test account
- Verify the anomaly is captured in the insider risk dashboard
- Expected Result: Anomalous activity generates a risk signal
- Evidence: Insider risk alert or risk score increase for test account
Test 5: Data Risk Graph Accessibility
- Objective: Confirm data risk graphs are available and integrated into investigation procedures
- Steps:
- Navigate to Microsoft Purview > Insider Risk Management > Investigations
- Access the Data risk graphs view
- Select a time window and verify graph data is loading
- Confirm the graph includes Copilot and agent interaction data
- Verify investigators know to include graph review in the standard investigation procedure
- Expected Result: Data risk graphs accessible and displaying Copilot/agent interaction data
- Evidence: Screenshot of data risk graph showing activity data
Test 6: IRM Triage Agent Operation
- Objective: Verify the Triage Agent is functioning and producing useful context summaries
- Steps:
- Navigate to Microsoft Purview > Insider Risk Management > Alerts
- Select an active alert and review the Triage Agent context summary
- Verify the summary includes: activity type detected, data involved, user risk history context
- Confirm the alert severity categorization is reasonable
- For Regulated tier: verify that the human-in-the-loop workflow requires investigator confirmation before alert dismissal
- Expected Result: Triage Agent produces actionable context summaries; severity categorizations are appropriate
- Evidence: Screenshot of Triage Agent context summary for an active alert
Test 7: Alert Triage Workflow
- Objective: Verify the alert triage process functions correctly
- Steps:
- Identify an active insider risk alert (or create one via testing)
- Verify the alert is routed to the assigned investigator
- Complete the triage workflow: review Triage Agent context, classify, and take action
- For agent risk alerts: verify routing reaches both compliance and agent deployment owners
- Verify the triage is documented in the case management system
- Expected Result: Alert is triaged per the documented workflow with Triage Agent context reviewed
- Evidence: Completed triage record with investigator actions
Test 8: Privacy Controls Verification
- Objective: Confirm pseudonymization and privacy controls are active
- Steps:
- Navigate to Insider Risk Management as an investigator
- Verify user identities are pseudonymized in the initial alert view
- Verify de-pseudonymization requires appropriate authorization
- Confirm privacy settings match the documented configuration
- Expected Result: Privacy controls function as configured
- Evidence: Screenshot showing pseudonymized user identities
Evidence Collection
| Evidence Item |
Format |
Storage Location |
Retention |
| Insider risk policy configuration |
PDF |
Compliance evidence repository |
7 years |
| Risky Agents policy configuration |
PDF |
Compliance evidence repository |
7 years |
| AI usage indicator configuration |
Screenshot |
Compliance evidence repository |
7 years |
| Triage Agent model inventory entry |
PDF |
Model risk management repository |
7 years |
| Alert and triage records |
PDF |
Compliance evidence repository |
7 years |
| Usage anomaly reports |
CSV |
Compliance evidence repository |
7 years |
| Privacy control verification |
Screenshot |
Compliance evidence repository |
7 years |
| Data risk graph samples |
Screenshot |
Compliance evidence repository |
7 years |
Compliance Mapping
| Regulation |
Requirement |
How This Control Supports It |
| FINRA Rule 3110 |
Supervisory systems |
Insider risk detection supports compliance with supervisory monitoring requirements; Risky Agents addresses the 2026 FINRA Oversight Report requirement for AI agent supervisory controls |
| FINRA Rule 3120 |
Testing supervisory procedures |
IRM alert generation and investigation workflow provide testable supervisory procedures for Copilot and agent oversight |
| FINRA 2026 Oversight Report (GenAI) |
Agentic AI supervisory controls |
Risky Agents policy and AI usage indicators directly address FINRA's 2026 requirement for supervisory systems covering AI workflow engines |
| OCC Bulletin 2025-26 |
Model risk management |
IRM Triage Agent documented as model per SR 11-7; proportionate AI-assisted governance |
| SEC Regulation S-P |
Safeguards for customer data |
Detecting data theft patterns helps meet customer data protection obligations |
| GLBA 501(b) |
Monitoring and testing safeguards |
IRM provides ongoing monitoring evidence and quarterly testing capability |
| Bank Secrecy Act |
Suspicious activity monitoring |
Copilot and agent usage monitoring supports compliance with suspicious activity reporting |
| NIST CSF |
DE.AE-1 Anomaly detection |
Insider risk supports compliance with anomaly detection requirements |