Skip to content

Control 1.6 — Verification & Testing: DSPM for AI

Control: 1.6 Microsoft Purview DSPM for AI Reference template: 1.7/verification-testing.md, 2.1/verification-testing.md Last UI Verified: April 2026

Re-verification cadence (SR 11-7 alignment)

Cadence Activity Owner
Daily (Zone 3) Activity Explorer review for in-scope users; oversharing remediation aging AI Governance Lead delegate
Weekly Default Weekly Risk Assessment review (top 100 sites); deterministic 1.6-ACT-01 test Compliance Admin
Monthly License entitlement reconciliation (1.6-LIC-01); one-click policy inventory (1.6-POL-01) Compliance Admin
Quarterly Negative test suite (1.6-NEG-01…1.6-NEG-05); attestation pack Compliance Admin + AI Governance Lead
On-change Re-run any test affected by license change, role change, policy change, sovereign-cloud change Change requester
On-incident 1.6-INC-01 — full evidence preservation per troubleshooting playbook Incident commander

Test catalog

Each test specifies prerequisites, deterministic input/output, and audit-binder evidence.

1.6-ACT-01 — Deterministic Copilot interaction is logged

Prereq: A named M365 Copilot–licensed user; tenant in scope of a Capture interactions for Copilot experiences template with content capture on.

Step Action Expected Evidence
1 Record UTC timestamp — tester log
2 User issues a known prompt referencing a labeled doc — tester log
3 Wait 24 h (per Learn) — —
4 Activity Explorer: filter User=<UPN> AND time window Event count ≥ 1, app=Microsoft 365 Copilot CSV export + SHA-256
5 Search-UnifiedAuditLog -RecordType CopilotInteraction over same window Same event count and UserId CSV export + SHA-256

Pass: Both steps 4 and 5 return ≥ 1 with reconciled count. Fail (silent-zero-row): Either returns 0 — investigate license, audit ingestion (correct shell), scope, content capture.

1.6-LIC-01 — License entitlement coverage

Pass: 100% of in-scope monitored users carry M365 Copilot SKU per Get-MgUserLicenseDetail; PAYG billing active where non-MS AI in scope. Fail: Any unlicensed in-scope user — silent under-reporting risk.

1.6-POL-01 — One-click policy inventory + content-capture state

Pass: Each enabled template has recorded {name, mode, scope, exclusions, content-capture state, role used}. Templates with "Capture" in name have content capture on. Fail: Any "Capture …" template with content capture off — Activity Explorer rows render but content is empty.

1.6-LBL-01 — Sensitivity-label propagation to Copilot response

Label a source doc → invoke Copilot summarization → assert response carries label or restriction per Copilot label inheritance behavior. (Per Learn microsoft-365-copilot-architecture-data-protection-auditing.)

1.6-AP-01 — Adaptive Protection threshold fires (Commercial / GCC only)

Induce a user into the elevated risk tier (test tenant) → attempt sensitive prompt → assert configured DLP action (warn / audit / block) fires and Activity Explorer + IRM both reflect the event. Skip on GCC High / DoD with documented exception.

1.6-WRA-01 — Weekly Risk Assessment cadence

Check Expected Source
Default Weekly Assessment runs each week Yes Data risk assessments page
First-results delay tolerance ≤ 4 days Learn dspm-for-ai-considerations
Refresh tolerance ≥ 48 h post-completion Learn
Custom assessments scheduled for Zone 3 sites > top 100 All in-scope sites covered CAB-tracked register

1.6-NEG-01..05 — Negative tests

ID Scenario Expected
NEG-01 Unauthorized role opens DSPM for AI Access denied
NEG-02 One-click policy paused (Mode=Disable) → test interaction Event logged but no enforcement; documented
NEG-03 Browser extension absent on managed Windows endpoint → ChatGPT visit Third-party AI event not in Activity Explorer
NEG-04 Unlicensed Copilot user attempts interaction Documented behavior (event still in audit vs. dropped)
NEG-05 Restricted-AU admin attempts to create one-click DSPM policy Refused (AU not supported)

1.6-WORM-01 — Evidence integrity & immutable storage

Pass: Every CSV / JSON / PDF artifact has a paired .sha256 sidecar; storage location is immutable (Purview retention label, SharePoint hold, or WORM blob) with retention aligned to Control 1.7. Fail: Any artifact without sidecar or stored on writable share.

1.6-DSPMv-01 — Unified DSPM (preview) accessibility

If tenant is opted into the preview, verify the DSPM (preview) node loads and Posture / Objectives / AI observability / Discover > Activity explorer / Discover > Data risk assessments are reachable. If not opted in, document and skip — do not assert preview-specific UI affordances.

Evidence pack (audit-binder)

Control-1.6_{TenantId}_{Cloud}_{ArtifactType}_{YYYYMMDD-HHmm-UTC}.{ext}
Control-1.6_{TenantId}_{Cloud}_{ArtifactType}_{YYYYMMDD-HHmm-UTC}.{ext}.sha256

Required artifacts per quarter:

  • 1.6-ACT-01 deterministic-test result (CSV) + tester log (TXT)
  • 1.6-LIC-01 license entitlement reconciliation (CSV)
  • 1.6-POL-01 one-click policy inventory (JSON)
  • 1.6-LBL-01 propagation test result (PDF + screenshots)
  • 1.6-AP-01 Adaptive Protection result (CSV + IRM event export) or documented sovereign exception
  • 1.6-WRA-01 weekly assessment summaries × 13 weeks (PDF + CSV)
  • 1.6-NEG-01..05 negative-test results (CSV + screenshots)
  • 1.6-WORM-01 immutable-storage attestation (signed)
  • Tenant cloud + role-by-step attestation
  • PowerShell transcripts from all runs

Attestation template

Control 1.6 — Microsoft Purview DSPM for AI
Quarter: Q_____ FY_____
Tenant: __________________________  Cloud: ☐ Commercial ☐ GCC ☐ GCC High ☐ DoD  Zone: ☐ 1 ☐ 2 ☐ 3

I have executed the test catalog above for the period covered. The evidence pack referenced
in this attestation supports — but does not by itself establish — the firm's compliance with:

  • FINRA Rule 3110 / 25-07 supervisory-system requirements applicable to AI surfaces
  • SEC 17a-4(f) / FINRA 4511 record-preservation expectations (paired with Audit Premium / Control 1.7)
  • SEC Reg S-P §248.30(a)(4) detection support for events that may trigger customer notification
  • GLBA 501(b) safeguards expectations for customer information processed by AI
  • OCC 2011-12 / Fed SR 11-7 model risk management ongoing-monitoring expectations
  • Interagency Guidance on Third-Party Relationships (OCC/FRB/FDIC) for ongoing monitoring of third-party AI

This evidence does not constitute a legal determination. Reportability decisions remain with
Compliance and Legal counsel.

Reviewer: __________________________   Role: __________________________
Signature: _________________________   Date (UTC): ____________________

Cross-references

Updated: April 2026 | Version: v1.4.0 | UI Verification Status: Current