Skip to content

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 the required Generative AI apps indicators and any Risky AI usage indicators (preview) in scope
  • Expected Result: All configured insider risk policies are active with the required generative-AI indicators enabled
  • Evidence: Policy status screenshot and PowerShell output

Test 2: Risky Agents Policy Verification

  • Objective: Confirm the Risky Agents (preview) policy is active and covers all deployed agent types
  • Steps:
  • Navigate to Microsoft Purview > Insider Risk Management > Policies
  • Locate the default Risky Agents policy
  • Verify the policy scope includes all deployed Copilot Studio, Microsoft Foundry, and P4AI SDK agents
  • Confirm alert routing is configured to reach both compliance and agent deployment owners
  • If the default policy doesn't meet the organization's needs, verify a custom policy exists for the additional FSI requirements instead of assuming the default policy is editable
  • Expected Result: Risky Agents policy is active, scoped correctly across all supported agent types, and alert routing is configured; any extra FSI tuning is implemented through supported custom-policy paths
  • Evidence: Policy configuration screenshot; alert routing configuration; custom policy evidence if applicable

Test 3: Generative AI Indicator Functionality

  • Objective: Verify the generative-AI monitoring indicators are active and producing signals
  • Steps:
  • Navigate to Microsoft Purview > Insider Risk Management > Settings > Policy indicators
  • Confirm the Generative AI apps indicators section is configured and the applicable Risky AI usage indicators (preview) are enabled
  • Generate test activity that matches the documented signal types, such as risky prompts, sensitive responses, or approved companion indicators
  • Wait for the processing cycle (24-48 hours)
  • Check the test user's risk timeline for risky AI usage or other configured AI-related signals
  • Expected Result: AI-related indicators appear in the risk timeline for elevated-activity accounts
  • Evidence: IRM risk timeline showing the triggered AI-related signals

Test 4: Anomaly Detection Functionality

  • Objective: Verify the system detects anomalous Copilot usage patterns
  • Steps:
  • Generate a test scenario that should trigger the configured IRM policy, such as a risky AI prompt, sensitive AI response, or approved companion exfiltration signal
  • 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: The configured risky-AI or companion signal generates a risk signal
  • Evidence: Insider risk alert or risk score increase for test account

Test 5: Data Risk Graph Accessibility

  • Objective: Confirm the data risk graph is set up and integrated into investigation procedures
  • Steps:
  • If the tenant previously onboarded before September 24, 2026, navigate to Microsoft Purview > Insider Risk Management > Recommended actions and confirm the Set up data lake and data risk graph action shows Complete
  • Navigate to Insider Risk Management > Alerts (preview) and open an alert
  • Select the Data risk graph tab and verify graph data is loading (allow 24-48 hours after initial onboarding for data to populate)
  • Confirm the graph displays the alert-related SharePoint/OneDrive exfiltration activity (sharing links, downloads, renames) for the 30-day window
  • Verify investigators know to include data risk graph review in the standard investigation procedure for alerts involving potential cross-department data movement
  • Expected Result: For previously onboarded tenants, the data risk graph is set up and accessible from alerts while the feature remains available, displaying alert-related exfiltration activity
  • Evidence: Screenshot of the data risk graph tab 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 > Agents > Explore agents and confirm the Insider Risk Management Triage Agent is deployed
  • Navigate to Microsoft Purview > Insider Risk Management > Alerts (preview)
  • 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) in Alerts (preview)
  • 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
Generative AI / Risky 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 the configured Generative AI apps / Risky 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
- Back to Control 2.10