Control 2.2: Sensitivity Labels and Copilot Content Classification — Verification & Testing
Test cases and evidence collection for validating sensitivity label enforcement with Copilot. This playbook covers verification for label groups, Copilot Studio agent label inheritance, and nested auto-labeling conditions.
Test Cases
Test 1: Label Taxonomy Structure Verification (Label Groups)
- Objective: Confirm the label taxonomy has been migrated to or is configured correctly with label groups (GA January 2026)
- Steps:
- Run Script 5 (Label Groups Migration Status Check) to inventory current label structure
- Verify all labels are organized within the expected label groups (Public, Internal, Confidential, Highly Confidential)
- If parent/child hierarchy is still in use, confirm migration is scheduled and document the timeline
- After migration: confirm all DLP policies that referenced parent label names continue to function by testing with labeled content
- Expected Result: Label taxonomy is organized in label groups; DLP policies verified to function correctly with new structure
- Evidence: Label taxonomy export (Script 5 output) and DLP policy test results post-migration
Test 2: Label Inheritance from Source Documents
- Objective: Verify Copilot-generated content inherits the appropriate sensitivity label from source documents
- Steps:
- Create a test document labeled "Confidential — Standard" in SharePoint
- Use Copilot to create a summary of the Confidential document
- Check the sensitivity label on the Copilot-generated summary
- Verify it inherits the "Confidential" label (or higher)
- Expected Result: Generated content inherits the source document's sensitivity label
- Evidence: Screenshots showing source label and generated content label
Test 3: Copilot Studio Agent Label Inheritance Verification
- Objective: Confirm Copilot Studio agents inherit the highest sensitivity label from their knowledge sources
- Steps:
- Identify a deployed Copilot Studio agent with knowledge sources carrying different sensitivity labels
- Document the labels on each knowledge source (e.g., "Confidential — Standard" and "Internal — General")
- Interact with the agent and confirm DLP monitoring triggers at the appropriate label tier (Confidential in this example)
- Verify the agent deployment record documents the inherited label
- For a new agent with an unlabeled knowledge source: add a labeled document to the knowledge base and verify the inherited label updates
- Expected Result: Agent's effective inherited label matches the highest label across all knowledge sources; DLP policies respond at the appropriate tier
- Evidence: Agent knowledge source label inventory and DLP monitoring evidence
Test 4: Auto-Labeling with Nested Conditions
- Objective: Verify nested AND/OR/NOT auto-labeling conditions apply labels correctly to FSI financial content
- Steps:
- Create a test document that matches a nested condition (e.g., contains both a CUSIP pattern AND an earnings context keyword)
- Upload the document to a SharePoint site covered by the auto-labeling policy
- Verify the policy applies the expected label (Confidential — MNPI per the nested rule)
- Create a second test document matching only part of the nested condition (e.g., CUSIP pattern only, no earnings keyword) and verify the label is NOT applied
- Create a third test document that would match conditions but is in an excluded folder (NOT condition) and verify the label is NOT applied
- Expected Result: Nested auto-labeling conditions correctly apply and withhold labels based on combined condition logic
- Evidence: Screenshots of labeled and unlabeled test documents with condition logic documentation
Test 5: Default Label Application
- Objective: Confirm default labels are applied to new Copilot-created content
- Steps:
- Use Copilot to draft a new document from scratch (no source documents)
- Check the sensitivity label on the new document
- Verify the default label ("Internal — General" or configured default) is applied
- Repeat across Word, Excel, and PowerPoint
- Expected Result: Default sensitivity label applied to all new Copilot content
- Evidence: Screenshots of new documents with default labels
Test 6: Mandatory Labeling Enforcement
- Objective: Verify users cannot save Copilot content without a label
- Steps:
- Have Copilot generate a document
- Attempt to remove the sensitivity label
- Attempt to save the document without a label
- Verify the system prevents saving and prompts for label selection
- Expected Result: System prevents saving unlabeled content
- Evidence: Screenshot of mandatory labeling prompt
Test 7: Label Downgrade Justification
- Objective: Confirm label downgrades require justification and are audited
- Steps:
- Open a document labeled "Confidential — Standard" that was created by Copilot
- Attempt to change the label to "Internal — General" (downgrade)
- Verify a justification prompt appears
- Enter a justification and complete the downgrade
- Verify the justification is recorded in the audit log (run Script 2 to confirm)
- Expected Result: Downgrades require justification; justification is audited
- Evidence: Justification prompt screenshot and audit log entry
Evidence Collection
| Evidence Item |
Format |
Storage Location |
Retention |
| Label taxonomy export (label groups migration status) |
CSV |
Compliance evidence repository |
7 years |
| Label inheritance test results |
PDF with screenshots |
Compliance evidence repository |
7 years |
| Agent label inheritance inventory |
PDF |
Compliance evidence repository |
7 years |
| Nested auto-labeling condition test results |
PDF with screenshots |
Compliance evidence repository |
7 years |
| Default label verification |
Screenshots |
Compliance evidence repository |
7 years |
| Mandatory labeling test |
Screenshots |
Compliance evidence repository |
7 years |
| Label event audit logs |
CSV |
Compliance evidence repository |
7 years |
Compliance Mapping
| Regulation |
Requirement |
How This Control Supports It |
| FINRA Rule 3110 |
Information classification |
Label enforcement supports compliance with content classification requirements; agent label inheritance extends classification to AI-generated content |
| SEC Regulation S-P |
Customer information protection |
Labels help classify and protect AI-generated content containing NPI; nested auto-labeling enables precise classification of complex financial data |
| GLBA Safeguards Rule |
Data classification controls |
Sensitivity labels and label groups provide systematic data classification for AI outputs and Copilot interactions |
| NIST CSF |
PR.DS-1 Data at rest protection |
Labels enforce protection policies on stored Copilot content; agent inherited labels extend protection to agent interactions |