Control 1.6 — Troubleshooting: DSPM for AI
Control: 1.6 Microsoft Purview DSPM for AI
Last UI Verified: April 2026
§1 — FSI Incident Handling — READ FIRST
DSPM for AI is the detection plane for Copilot / agent / third-party-AI risk. A loss or degradation of DSPM telemetry on a Zone 3 workload is a supervisory-system gap under FINRA 3110 and a books-and-records integrity event under FINRA 4511 / SEC 17a-4(f). Treat as an incident; do not "just fix the UI."
Severity matrix
Severity
Trigger
Response window
Escalation
SEV-1
DSPM telemetry blackout on a Zone 3 workload affecting M365 Copilot or named regulated agents; sensitive-info exposure visible without policy enforcement; unauthorized access to DSPM Content Viewer
Immediate
CISO + Compliance + Legal within 1 h
SEV-2
Degraded coverage (one Get Started step regressed; one-click template not enforcing); Adaptive Protection inactive; missing PAYG coverage of in-scope third-party AI
4 h
Compliance Admin → AI Governance Lead within 4 h
SEV-3
Single-user / single-site coverage gap; assessment delayed > documented Learn window
1 business day
Compliance Admin
SEV-4
Cosmetic / UI drift; preview feature regression
Best effort
Track in known-issues log
Reportability decision tree
Reportability is a Compliance/Legal determination. Use this tree to escalate, not to decide.
Trigger
Escalate to
Possible obligation
Loss of supervisory visibility on AI-surfaced communications
Compliance
FINRA 3110 supervisory-system obligation
Books-and-records gap (CopilotInteraction events lost / not retained)
Compliance + Legal
FINRA 4511 / SEC 17a-4(f)
NPI / customer info disclosed via Copilot/agent without enforcement
Privacy + Legal
GLBA 501(b); SEC Reg S-P §248.30(a)(4) customer-notification timeline
Cybersecurity event materially affecting normal operations
CISO + Legal
NY DFS 23 NYCRR 500 — 72-hour determination
AI/model-related operational risk event
Model Risk + Compliance
OCC 2011-12 / Fed SR 11-7
Third-party AI vendor failure
Vendor mgmt + Compliance
Interagency Guidance on Third-Party Relationships
Records-related event for a covered swap/trading activity
Compliance
CFTC Rule 1.31
Insider misconduct involving AI surface
HR + Legal + Compliance
FINRA 4530
Capture, do not mutate first:
Screenshots of the failure state (DSPM Overview, the failing page, error toast)
Get-AdminAuditLogConfig output from Connect-ExchangeOnline (correct shell)
One-click policy inventory snapshot (Get-DlpCompliancePolicy, Get-RetentionCompliancePolicy)
Last successful evidence pack timestamp
Tenant ID, cloud, affected user/site list, UTC window, role used
Browser-extension and device-onboarding state
Activity Explorer export for the suspect window (paginated; no truncation)
Adaptive Protection / IRM policy snapshot
SHA-256 sidecar for every artifact
Compensating controls during the gap
Increase Communication Compliance review cadence (Control 1.10) on in-scope reviewers
Tighten DLP for Copilot location (Control 1.5) to Block where previously TestWithNotifications
Freeze new Zone 3 agent activations / new Copilot Studio publishes (Control 2.1, 2.16)
Manually search Unified Audit Log daily for CopilotInteraction events (Control 1.7)
Increase manual supervisory review of AI-assisted content for in-scope user populations
Pre-escalation checklist (≥ 12 items)
Tenant ID and cloud confirmed (Commercial / GCC / GCC High / DoD)
All four Get Started steps re-checked with screenshots
License floor verified via Graph (Get-MgSubscribedSku)
Per-user Copilot license verified for sample of affected users
PAYG billing state verified (where third-party AI in scope)
Audit ingestion verified from EXO session, not IPPS
One-click policy inventory captured with mode + content-capture state
Activity Explorer queried with deterministic filter; result count recorded
Mirror query run via Search-UnifiedAuditLog -RecordType CopilotInteraction (paginated)
Browser extension / Edge configuration policy state verified per device class
Device onboarding state verified (Defender for Endpoint / standalone)
Adaptive Protection / IRM policy state captured
Sovereign cloud parity confirmed (IRM/Adaptive Protection NOT at parity in GCC High/DoD)
Administrative-unit restriction ruled out
Last known good evidence pack timestamp recorded
Compliance + Legal notified per severity matrix
§2 — Symptom-driven diagnostics
Symptom: DSPM Activity Explorer shows zero AI interactions
Cause
Diagnostic
Fix
Audit ingestion off
Get-AdminAuditLogConfig from EXO session
Re-enable in EXO; do not trust IPPS value
Wrong RecordType used in PowerShell mirror query
Inspect script for AIPDiscover etc.
Use CopilotInteraction
Single-shot pagination truncated
Inspect for missing -SessionId / -SessionCommand ReturnLargeSet
Implement do-while pagination, cap 50,000 per session
Affected users unlicensed for M365 Copilot
Get-MgUserLicenseDetail
Assign Copilot license; document gap
Content capture not enabled on collection policy
DSPM > Policies > template > content capture state
Enable; allow up to 24 h to surface
Browser extension / Edge config policy missing on managed device
Intune device compliance report
Push policy / extension; reconcile coverage
Affected user without Exchange Online mailbox
Graph user lookup
Prompt/response text won't display per Learn
Restricted-AU admin opening the page
Confirm user's AU scope
Use a tenant-scoped reviewer; AU not supported
Symptom: One-click policy template not visible / "Recommendation Not Updating"
Cause
Fix
Missing role group (e.g., IRM role group for Risky AI usage)
Assign per template's underlying solution
License floor unmet (E5 / E5 Compliance / Purview Suite)
Assign SKU; reload page after propagation
Recommendation completed/dismissed/PendingDeletion
Check status; do not "manually mark complete" without evidence — that suppresses oversight
Restricted-AU admin (DSPM does not support AU)
Use tenant-scoped admin
GCC High/DoD parity gap
Document exception; use commercial-only feature only where the workload allows
Symptom: DSPM Get Started never completes / "Activate Audit" step stuck
Validate UnifiedAuditLogIngestionEnabled = True from EXO session
If True but UI cached False, sign out / refresh token; allow propagation
Confirm an Exchange role group holder is performing the step (Compliance Admin alone cannot)
"Extend your insights" requires IRM / Communication Compliance pre-reqs — incomplete pre-reqs leave the step "Completed" with degraded observability
Symptom: Adaptive Protection score does not change
Requirement
Verification
IRM policy created and active
Get-InsiderRiskPolicy (where exposed) / portal
Minimum baseline window observed
≥ 7 days per Learn
Risk-level thresholds configured in IRM
Insider Risk > Settings
User in scope of IRM policy
IRM policy scope
Tenant cloud supports IRM/Adaptive Protection
Skip in GCC High/DoD; document
Symptom: Sensitivity label not propagated to Copilot response
Cause
Fix
Endpoint MIP client out of date
Update to supported client version
Label not published to user
Publish via Sensitivity labels > Publish
Auto-labeling scope excludes the file
Review scope conditions
Container-label inheritance off
Enable per Learn sensitivity-labels-teams-groups-sites
DLP for Copilot location does not allow SIT + label in same rule
Split into separate rules per Learn
Symptom: Third-party AI (ChatGPT Enterprise / Gemini) shows no events
Cause
Fix
PAYG billing not enabled
Enable Purview pay-as-you-go in Azure subscription
Browser extension missing on Windows + Chrome/Firefox
Push via Intune
Edge configuration policy missing
Push via Intune (NOT extension)
Device not onboarded
Defender for Endpoint or standalone Purview onboarding
Extended insights step not complete (requires IRM pre-reqs)
Complete Get Started step 4
§3 — DSPM vs Audit decision matrix
Question
Use DSPM Activity Explorer
Use Search-UnifiedAuditLog (Control 1.7)
What was the prompt content / response category?
✅ (with content viewer role + content capture on)
⚠️ Partial via CopilotInteraction.AuditData.CopilotEventData
Books-and-records evidence for FINRA 4511 / SEC 17a-4(f)
❌ (UI tool, ~30-day window, not WORM)
✅ (paired with Audit Premium retention)
Third-party AI app interaction (ChatGPT / Gemini)
✅ (browser extension + PAYG)
❌ (not captured in UAL)
Long-horizon (> 180 d) lookback
❌
✅ (Audit Premium + retention policy)
Per-agent risk score
✅ (preview)
❌
§4 — Anti-patterns (do not do)
Anti-pattern
Why it's wrong
"Manually mark recommendation as complete" without evidence
Suppresses oversight without proving the underlying control is effective
Treating Activity Explorer empty table as PASS
Silent-zero-row trap (N3.2) — assert deterministic event
Filtering audit search with AIPDiscover etc. for Copilot evidence
AIP record types ≠ DSPM-for-AI events
Calling Set-AdminAuditLogConfig from Connect-IPPSSession
Wrong-shell trap (N3.1)
Asserting RiskScore / AccessPattern fields on CopilotInteraction.AuditData
Not in published schema
Using commercial portal URL on a GCC High tenant
purview.microsoft.com on a .us tenant returns wrong scope
Assuming Adaptive Protection works in GCC High / DoD
IRM not at parity
Restricted-AU admin attempting one-click policy creation
AU not supported by DSPM
§5 — Escalation path
L1 — Compliance Admin (within 1 h for SEV-1; 4 h for SEV-2): preserve evidence, run pre-escalation checklist
L2 — AI Governance Lead (within 1 h for SEV-1): triage cross-control impact (1.5 / 1.7 / 1.10 / 1.12 / 2.1)
L3 — CISO + Compliance Officer + Legal (within 1 h for SEV-1): reportability determination
L4 — Microsoft support with: tenant ID, cloud, affected workload, UTC window, evidence pack reference, severity, business impact statement
L5 — Regulator notifications as determined by Legal / Compliance
Cross-references
Updated: April 2026 | Version: v1.4.0 | UI Verification Status: Current