Skip to content

Control 2.1: DLP Policies for M365 Copilot Interactions — Verification & Testing

Test cases and evidence collection for validating DLP policy effectiveness for Copilot. This playbook covers verification for label-based response blocking, SIT-based prompt blocking, SIT-based web-search restriction, and the default Microsoft-deployed policy.

Test Cases

Test 1: DLP Policy Activation Verification (All Copilot DLP Rule Types)

  • Objective: Confirm all Copilot DLP rule types are active and properly configured
  • Steps:
  • Run Script 3 to list all Copilot DLP policies
  • Verify at least one label-based response blocking policy exists and is in the expected mode (Test or Enforce)
  • Verify at least one SIT-based prompt blocking policy exists and is in the expected mode
  • Verify at least one SIT-based web-search restriction policy exists and is in the expected mode
  • Locate the Microsoft-deployed default Copilot DLP policy (simulation mode) and confirm it is present
  • Confirm policy locations include Microsoft 365 Copilot as the monitored location
  • Verify rule conditions match the approved DLP strategy for each rule type
  • Expected Result: All Copilot DLP rule types and the default policy are present; each is in the expected mode
  • Evidence: DLP policy configuration export showing all Copilot DLP rules and the default policy

Test 2: Label-Based Response Blocking (Type 1 Verification)

  • Objective: Verify the label-based policy blocks Copilot from surfacing Highly Confidential content
  • Steps:
  • Upload a test document labeled "Highly Confidential" to a SharePoint site accessible to a test user
  • As the test user, ask Copilot to summarize or reference the test document
  • Verify Copilot does not include the document content in its response
  • Capture the user message shown, if any; do not require a classic DLP policy tip because Copilot can instead display an organizational-policy message
  • Confirm the match event is logged in the DLP incident report
  • Clean up test data after verification
  • Expected Result: Copilot is blocked from surfacing the Highly Confidential document; event is logged
  • Evidence: Screenshot showing the excluded-content behavior and any user message, plus the DLP incident record

Test 3: SIT-Based Prompt Blocking (Type 2 Verification)

  • Objective: Verify the SIT-based policy blocks Copilot when the user types sensitive data in a prompt
  • Steps:
  • With the SIT-based prompt blocking policy active, type a prompt containing a test SSN pattern (000-00-0000) directly into Copilot
  • Verify Copilot does not process the request — Copilot should be blocked from responding
  • Capture the user message shown. During preview, Word, Excel, and PowerPoint might not clearly identify organizational policy as the cause even when the response is blocked.
  • Confirm the block event is recorded in DLP incident reports
  • Verify the enforcement happens on the prompt side (before Copilot retrieves any content)
  • Expected Result: Copilot is blocked from responding to a prompt containing a sensitive data pattern
  • Evidence: Copilot response showing blocked prompt and DLP incident record

Test 3b: SIT-Based Web Search Restriction (Type 3 Verification)

  • Objective: Verify the SIT-based web-search restriction blocks external web search when prompt text contains configured sensitive information types
  • Steps:
  • With the SIT-based web-search restriction policy active, type a prompt containing a test SSN pattern (000-00-0000) and a request that would otherwise use external web search
  • Verify Copilot blocks external web search as a grounding source for that prompt
  • Verify Copilot may still answer using permitted internal Microsoft 365 data sources
  • Confirm the match event is recorded in DLP incident reports
  • Capture evidence that the enforcement point is the external web-search grounding path, not a complete prompt-processing block
  • Expected Result: External web search is blocked for a sensitive prompt, while permitted internal grounding remains available
  • Evidence: Copilot response or admin evidence showing web-search restriction and DLP incident record

Test 3c: Direct Prompt Upload Coverage Gap

  • Objective: Confirm the control design does not incorrectly claim that Copilot DLP scans the contents of files uploaded directly with a prompt
  • Steps:
  • Review the current Microsoft Learn limitation for files uploaded in prompts
  • Confirm the DLP test plan distinguishes typed prompt text from uploaded file content
  • Verify complementary controls are documented for direct uploads, such as approved storage, sensitivity labels, endpoint controls, and user guidance
  • Expected Result: The organization does not rely on SIT prompt inspection to scan directly uploaded file contents
  • Evidence: Approved control-gap statement and complementary-control mapping

Test 4: Default Policy Simulation Mode Review

  • Objective: Validate that the Microsoft-deployed default Copilot DLP policy is capturing expected matches
  • Steps:
  • Access the default policy via MAC > Copilot > Security or Purview > DLP > Policies
  • Review simulation mode match data for the past 14+ days
  • Classify a sample of 20 matches as true positive or false positive
  • Calculate the false positive rate
  • If false positive rate is below 10%, document readiness to transition from simulation to enforcement
  • Expected Result: Simulation data captured; false positive rate documented; enforcement readiness assessed
  • Evidence: Simulation match data export and false positive analysis

Test 5: Block Action Enforcement

  • Objective: Confirm DLP blocking actions prevent sensitive data from being exposed through Copilot across all rule types
  • Steps:
  • With all DLP rules in enforcement mode, repeat the label-based, SIT prompt-blocking, and SIT web-search restriction tests
  • Verify Copilot does not include sensitive labeled content in responses (Type 1), does not process sensitive prompts where Type 2 applies, and does not use external web search where Type 3 applies
  • Capture the user message shown for each supported client; do not require identical policy-tip wording across clients
  • Confirm block events are recorded separately by rule type in DLP incident reports
  • Expected Result: All Copilot DLP rule types block as designed; events recorded
  • Evidence: Copilot responses showing blocked content or blocked web search and DLP incident records for each rule type

Test 6: Edge Browser DLP Coverage

  • Objective: Verify DLP policies apply to Copilot interactions accessed through Microsoft Edge browser
  • Steps:
  • Access Copilot via Microsoft Edge (browser-based access, e.g., m365copilot.com)
  • Type a prompt containing a test SSN pattern (000-00-0000); do not use an attached file as the SIT test input
  • Verify the Edge browser DLP policy triggers (audit or block)
  • Confirm the event appears in DLP incident reports attributed to the Endpoint DLP / Edge channel
  • Expected Result: Edge browser DLP applies to browser-based Copilot interactions
  • Evidence: DLP match event attributed to Edge/Endpoint DLP channel

Test 6a: Mac Endpoint DLP Expanded File Type Coverage

  • Objective: Verify Mac endpoint DLP covers expanded file types (100+ supported file types)
  • Steps:
  • On a managed Mac device with Endpoint DLP configured, create a test file in one of the newly supported file types containing a SIT pattern (e.g., a CSV or Numbers file with test SSN data).
  • Attempt to upload or share the file through a Copilot-accessible location.
  • Verify the Endpoint DLP policy triggers for the expanded file type.
  • Confirm the event appears in DLP incident reports.
  • Expected Result: Mac Endpoint DLP detects sensitive data in expanded file types.
  • Evidence: DLP match event for the expanded file type on a Mac device.

Test 6b: Adaptive Scoping for SharePoint DLP

  • Objective: Verify adaptive scoping dynamically applies DLP policies to SharePoint sites used as Copilot grounding sources
  • Steps:
  • Configure an adaptive scope targeting SharePoint sites with a specific sensitivity label (e.g., "Confidential").
  • Attach the adaptive scope to a DLP policy covering the SharePoint location.
  • Apply the target sensitivity label to a test SharePoint site and upload content containing a SIT pattern.
  • Verify the DLP policy applies to the newly scoped site without manual policy updates.
  • Remove the sensitivity label from the test site and verify the DLP policy no longer targets it.
  • Expected Result: Adaptive scoping dynamically includes and excludes SharePoint sites based on site properties.
  • Evidence: DLP policy match data showing dynamic site inclusion.

Test 7: Override and Justification Flow

  • Objective: Verify override mechanisms work correctly for permitted override scenarios
  • Steps:
  • Configure a test DLP rule with override allowed (business justification required)
  • Trigger the DLP policy with medium-sensitivity test content
  • Attempt to override with a business justification
  • Verify the justification is recorded in the audit log
  • Verify the override is granted and content is accessible
  • Expected Result: Override flow works correctly with justification recorded
  • Evidence: Override audit log entry with justification text

Test 8: False Positive Rate Assessment

  • Objective: Evaluate the DLP policy false positive rate to validate confidence level settings
  • Steps:
  • Run Script 4 to export DLP incidents from the past 30 days
  • Review a sample of 50 incidents and classify as true positive or false positive (assess each policy type separately)
  • Calculate the false positive rate per policy type
  • If rate exceeds 20% for either type, adjust confidence levels or SIT patterns
  • Expected Result: False positive rate below 20% for each policy type
  • Evidence: Incident classification analysis with false positive calculation per policy type

Evidence Collection

Evidence Item Format Storage Location Retention
DLP policy configuration export (all Copilot rule types + default) CSV/JSON Compliance evidence repository 7 years
Label-based response blocking test results PDF with screenshots Compliance evidence repository 7 years
SIT-based prompt blocking test results PDF with screenshots Compliance evidence repository 7 years
SIT-based web-search restriction test results PDF with screenshots Compliance evidence repository 7 years
Direct prompt upload gap review PDF/control mapping Compliance evidence repository 7 years
Default policy simulation review PDF with match data Compliance evidence repository 7 years
Edge browser DLP test results PDF with screenshots Compliance evidence repository 7 years
Mac endpoint DLP expanded file type test results PDF with screenshots Compliance evidence repository 7 years
Adaptive scoping test results PDF with match data Compliance evidence repository 7 years
DLP incident reports CSV Compliance evidence repository 7 years
False positive analysis (per policy type) PDF Compliance evidence repository 7 years

Compliance Mapping

Regulation Requirement How This Control Supports It
SEC Regulation S-P (17 CFR §248, amended Dec 3, 2025) Customer NPI safeguards covering AI interaction surfaces SIT-based prompt blocking (Type 2) addresses the requirement that customer information safeguards extend to AI interaction surfaces
FINRA Rule 3110 Supervisory data controls All Copilot DLP rule types support compliance with supervisory requirements for data protection in AI interactions
GLBA §501(b) Technical safeguards Copilot DLP rule types provide technical controls — Type 1 at the response layer, Type 2 at the prompt layer, and Type 3 at the external web-search grounding path
PCI DSS Cardholder data protection SIT-based prompt blocking prevents credit card data from entering Copilot prompts; SIT-based web-search restriction prevents configured sensitive prompt data from being sent to external web search providers; label-based blocking prevents labeled cardholder data from being surfaced in responses
- Back to Control 2.1