
Key Takeaways
- Microsoft Purview AI Governance & Compliance Guide (2026).
- TL;DR — Microsoft Purview AI Governance Components.
- Microsoft Purview DSPM for AI Day-1 Capabilities.
- Sensitivity-Label-Aware AI Grounding.
- DLP for Microsoft 365 Copilot Prompts and Grounding.
- Audit Retention for AI Interactions.
On this page13 sections
Microsoft Purview AI Governance & Compliance Guide (2026)
Microsoft Purview AI governance is the operational discipline of using Microsoft Purview DSPM for AI (formerly AI Hub), sensitivity labels, DLP, audit, and Compliance Manager to govern Microsoft 365 Copilot, Microsoft Power BI Copilot, Microsoft Copilot Studio agents, and Azure OpenAI custom applications across regulated industries.
EPC Group has delivered Microsoft Purview AI governance engagements for Fortune 500 healthcare, financial services, government, pharma, and EU-regulated organizations.
TL;DR — Microsoft Purview AI Governance Components
| Component | Purpose |
|---|---|
| Microsoft Purview DSPM for AI | Copilot activity reporting, oversharing assessment, one-click policies |
| Sensitivity Labels (Restricted-tier) | Block AI grounding on regulated content |
| DLP for Microsoft 365 Copilot | No answer for prompts with chosen sensitive info types; labeled items excluded from grounding |
| Microsoft Purview Audit (Premium) | 7-10 year retention for AI interactions |
| Microsoft Purview eDiscovery (Premium) | AI prompt history in litigation/regulatory scope |
| Microsoft Compliance Manager | AI-specific control attestation |
| Microsoft Sentinel | SOC monitoring for AI risk events |
Microsoft Purview DSPM for AI Day-1 Capabilities
Microsoft renamed AI Hub to Data Security Posture Management (DSPM) for AI and now also offers a newer Data Security Posture Management experience. Its Day-1 capabilities:
- Activity reporting — Copilot prompts and responses appear in the DSPM for AI Activity explorer for admins with the right permissions
- Oversharing assessment — reports on sensitive content that Copilot can reach through broad permissions
- One-click policies — ready-made DLP and Insider Risk Management policies for AI
- Risky prompts and responses — flagged by the Insider Risk Management Risky AI usage template rather than by DSPM for AI on its own
Sensitivity-Label-Aware AI Grounding
Microsoft Copilot grounding respects sensitivity labels:
| Label Tier | Microsoft Copilot Grounding Behavior |
|---|---|
| Public | Available for grounding |
| General | Available for grounding |
| Confidential | Available for grounding (logged) |
| Highly Confidential | Available for grounding (logged + risk-scored) |
| Restricted (PHI/MNPI/CUI) | BLOCKED from grounding |
The Restricted-tier block is the critical compliance gate. Documents labeled Restricted-PHI, Restricted-MNPI, Restricted-CUI never appear in Copilot grounding regardless of user permissions.
DLP for Microsoft 365 Copilot Prompts and Grounding
Microsoft Purview DLP for Microsoft 365 Copilot and Copilot Chat:
| Policy | Condition | Action |
|---|---|---|
| Stop sensitive prompts | SITs for SSN, PHI identifiers or MNPI keywords (custom SITs as needed) | Prevent Copilot from processing content, processing prompts (preview) |
| No web search for sensitive prompts | The same SITs | Prevent Copilot from processing content, performing web searches |
| Stop prompts carrying secrets | Credential SITs for API keys, tokens and passwords | Prevent Copilot from processing content, processing prompts (preview) |
| Keep labeled content out of answers | Sensitivity labels such as Restricted-PHI | Prevent Copilot from processing content (items can still be cited) |
| Keep outside email out of Copilot | Email received from external users (preview) | Prevent Copilot from processing content |
There is no response inspection or redaction and no prompt-injection or jailbreak condition. Label and SIT conditions go in separate rules, the location is offered only in the Custom template and switches off all other locations in the policy, and updates can take up to four hours to apply.
Audit Retention for AI Interactions
| Industry | Retention Required |
|---|---|
| HIPAA (healthcare) | 6 years (documentation) |
| FINRA Rule 4511 (financial) | 6 years |
| SEC Rule 17a-4 (broker-dealer) | 3 to 6 years (life of the enterprise for organizational records) |
| FedRAMP Moderate / High | 7 years |
| GxP (pharma) | 7+ years |
Microsoft Purview Audit (Premium) license + retention policy = compliance-grade AI audit posture.
Microsoft Compliance Manager AI Assessments
Built-in assessment templates for:
- EU AI Act — high-risk system documentation, transparency obligations, prohibited use cases, conformity assessment
- NIST AI RMF — Govern / Map / Measure / Manage functions
- ISO 42001 — AI management system
- HIPAA AI provisions — PHI handling in AI systems
- FINRA Rule 3110 supervision — supervised communications via AI
- GDPR Article 22 — automated decision-making with significant effect
Microsoft Sentinel Custom Analytics for AI
EPC Group standard custom analytics rule library:
// High-volume Restricted-tier grounding attempts
CopilotEvents
| where SensitivityLabel startswith "Restricted"
| where ResponseStatus == "Blocked"
| summarize attempts = count() by UserPrincipalName, bin(TimeGenerated, 1h)
| where attempts > 10
// Cross-tenant grounding (Information Barrier violation indicator)
CopilotEvents
| where GroundingScope has "cross-tenant"
| where SensitivityLabel in ("Confidential", "Highly Confidential", "Restricted")
// Off-hours / off-region Copilot usage
CopilotEvents
| where hourofday(TimeGenerated) !between (6 . 20)
| where DayOfWeek between (1 . 5)
| summarize off_hour_count = count() by UserPrincipalName
| where off_hour_count > 50
EU AI Act Compliance Mapping
For EU-regulated tenants:
Article 50 transparency obligations:
- User notice when interacting with AI system
- Clear identification of AI-generated content
- Logging of AI interactions for audit
High-risk system documentation:
- Risk assessment per use case
- Training data lineage
- Model evaluation results
- Human oversight documentation
- Continuous monitoring evidence
Microsoft Purview DSPM for AI supplies AI interaction records; risk assessment, documentation and human oversight remain customer work. Microsoft Compliance Manager EU AI Act assessment template tracks customer-side responsibility.
Microsoft Sentinel Integration Architecture
- DLP alerts, Activity explorer and DSPM for AI activity → documented sources for Copilot detections, each query tested in your own tenant
- Microsoft Sentinel → custom analytics rules + workbooks
- Microsoft Defender XDR → pre-correlated incidents
- ServiceNow / Jira → ticket automation via Logic Apps
- Microsoft Teams → SOC analyst notification
Pricing
| Component | Pricing |
|---|---|
| Microsoft 365 E5 | $60.00/user/month (Microsoft list, read 2026-10-21) |
Plan coverage for each Purview capability is listed in Microsoft's service description; confirm it against your own agreement.
EPC Group fixed-fee Microsoft Purview AI governance:
- Mid-market: fixed-fee
- Enterprise: fixed-fee
- Fortune 500: fixed-fee
Frequently Asked Questions
Is Microsoft Purview DSPM for AI mandatory for Copilot?
EPC Group treats it as mandatory for regulated industries (HIPAA, FINRA, SEC, FedRAMP, CMMC, GxP, EU AI Act) and strongly recommends it for non-regulated tenants. Day-1 enablement is the EPC Group standard.
How does Microsoft Purview integrate with Microsoft 365 Copilot?
Microsoft Copilot grounding respects Microsoft Purview sensitivity labels (Restricted-tier blocked). Copilot prompts and responses are reported in DSPM for AI. Microsoft Purview DLP acts on Copilot prompts and on the content Copilot may use, not on its responses. Microsoft Purview Audit captures all Copilot interactions.
What about EU AI Act compliance?
EU AI Act conformity assessment is supported by Microsoft Compliance Manager AI assessment template. DSPM for AI and Purview Audit supply interaction records as technical evidence. EPC Group EU AI Act guide covers the framework.
Who delivers Microsoft Purview AI governance engagements?
EPC Group senior architects with combined Microsoft Purview (since MIP era 2017+), Microsoft Defender, Microsoft Sentinel, and AI compliance experience. Errin O'Connor is a 4-time Microsoft Press & Sams author.
Next Steps
Schedule a 30-minute Microsoft Purview AI governance discovery call or call (888) 381-9725. Senior architects (not sales) take discovery calls.
Related reading: Microsoft Purview Data Governance Enterprise Guide, Microsoft Purview for Copilot Implementation, Microsoft Copilot Governance Framework for Regulated Industries, EU AI Act Microsoft Stack Implementation Guide, and NIST AI RMF Microsoft Stack Implementation Guide.
Errin O'Connor
Founder & Chief AI Architect
Microsoft Press bestselling author with enterprise consulting experience since 1997.
View Full ProfileRelated Articles
Silent AI Is Dead: What Six Insurance Carriers Told Me About Your 2026 Renewal
"Silent AI" ended January 1, 2026, when ISO generative-AI exclusions (CG 40 47/48) went live. Here is what six insurance carriers told me they now require before they will renew AI-touching coverage — and the four court cases driving it.
AI GovernanceMicrosoft Build 2026 for the Board: 5 Strategic Decisions for CIOs
A CIO board-prep framework for Build 2026 with the 5 strategic decisions that must land in Q3-Q4 2026: platform standardization, Agent 365, governance posture, compute budget, ROI measurement.
AI GovernanceMicrosoft Fabric Migration Risk: HIPAA, SOC 2, FedRAMP After Build 2026
Compliance risk assessment for Fabric migration after Build 2026: HIPAA controls, SOC 2 audit scope expansion, FedRAMP authorization gaps, EU AI Act implications, and the 14 controls regulated enterprises must add.
