Microsoft Purview DSPM for AI Review

Industrial cybersecurity

TL;DR

Last updated: November 3, 2026. Testing track: Documentation Reviewed.

Purview is Microsoft’s data security and compliance suite, and its DSPM experience is the console for AI apps and agents. It is strongest for Microsoft 365 Copilot, Copilot Studio, Foundry and Agent 365, and extends to third-party AI sites through Edge for Business, Endpoint DLP on onboarded devices, a browser extension and SASE integrations. Value depends heavily on existing Microsoft licensing, Intune management and sensitivity label maturity.

Ownership: Microsoft product (Nasdaq: MSFT). Not an acquisition. The original Purview AI Hub became DSPM for AI; in 2026 DSPM and DSPM for AI were merged into one DSPM experience and the old versions were renamed ‘classic’ (Microsoft Learn DSPM page, dated 2026-05-01). Message center MC1481315 says the classic experiences retire mid to late March 2027.

Score: 7.8 / 10.

Initial review November 3, 2026.

How This Review Was Conducted

This review is based on Microsoft Purview DSPM for AI’s public documentation, product pages, pricing pages and press coverage as of November 2026. We have not had lab access or a vendor demo.

Scores use the same published rubric and formula as every review on this site. Sources are listed at the end of the review.

Review methodology and evaluation process

Score breakdown

Dimension

Coverage breadth

Detection accuracy

Deployment friction

Policy & control depth

Framework alignment

Pricing transparency

Support & documentation

Weight

20%

20%

15%

15%

10%

10%

10%

Score

8

8

5

8

9

8

9

Notes

Broadest native coverage of Microsoft Copilot and agent surfaces plus enterprise AI connectors and third-party sites, but third-party browser coverage is Edge and Windows centric and MCP is not documented.

Mature SIT, classifier and label engine reused for AI, offset by documented gaps such as unscanned Copilot file uploads, size limits and encoded-traffic caveats, with no public benchmarks.

Copilot controls are tenant-native, but third-party AI enforcement requires device onboarding, Intune management, Edge enforcement and Conditional Access configuration.

Granular DLP with block, warn-override, labels, admin units and simulation, though third-party AI in Edge is limited to allow or block without redaction.

Documents EU AI Act, ISO/IEC 42001, ISO/IEC 23894 and NIST AI RMF templates, but no OWASP LLM or MITRE ATLAS mapping.

Per-user list prices are published on microsoft.com and AI meters are listed on Azure, though several meter rates are blank.

Extensive Microsoft Learn documentation and enterprise support, slightly hampered by frequent renames and preview churn.

Coverage breadth

Weight
20%

Score

8

Notes

Broadest native coverage of Microsoft Copilot and agent surfaces plus enterprise AI connectors and third-party sites, but third-party browser coverage is Edge and Windows centric and MCP is not documented.

Detection accuracy

Weight
20%

Score

8

Notes

Mature SIT, classifier and label engine reused for AI, offset by documented gaps such as unscanned Copilot file uploads, size limits and encoded-traffic caveats, with no public benchmarks.

Deployment friction

Weight
15%

Score

5

Notes

Copilot controls are tenant-native, but third-party AI enforcement requires device onboarding, Intune management, Edge enforcement and Conditional Access configuration.

Policy & control depth

Weight
15%

Score

8

Notes

Granular DLP with block, warn-override, labels, admin units and simulation, though third-party AI in Edge is limited to allow or block without redaction.

Framework alignment

Weight
10%

Score

9

Notes

Documents EU AI Act, ISO/IEC 42001, ISO/IEC 23894 and NIST AI RMF templates, but no OWASP LLM or MITRE ATLAS mapping.

Pricing transparency

Weight
10%

Score

8

Notes

Per-user list prices are published on microsoft.com and AI meters are listed on Azure, though several meter rates are blank.

Support & documentation

Weight
10%

Score

9

Notes

Extensive Microsoft Learn documentation and enterprise support, slightly hampered by frequent renames and preview churn.

What it does well

Native policy control over Microsoft 365 Copilot.

Native, policy-level control over Microsoft 365 Copilot, including blocking prompts with sensitive info types and excluding labeled files and external email from grounding (Microsoft Learn, Copilot DLP).

Publishes AI regulation templates in Compliance Manager.

Published AI regulation templates for EU AI Act, ISO/IEC 42001, ISO/IEC 23894 and NIST AI RMF in Compliance Manager (Microsoft Learn, Compliance Manager assessments).

Reuses existing labels and retention for AI.

Reuses existing sensitivity labels, SITs, Insider Risk, eDiscovery and retention for AI prompts and responses, so mature Purview customers add little new tooling (Microsoft Learn, Purview for AI).

Covers enterprise AI connectors and agents.

Covers enterprise AI connectors including ChatGPT Enterprise and Claude Enterprise, plus Foundry, Copilot Studio and Agent 365 agents (Microsoft Learn, Purview for AI). List prices for E5 and the Purview Suite are published, and AI pay-as-you-go meters are listed on the Azure pricing page (microsoft.com and azure.microsoft.com).

Where it falls short

Edge inline protection requires Intune-managed Windows devices.

Inline protection for unmanaged AI apps in Edge requires Intune-managed Windows devices and does not support tenants using Intune multi-admin approval (Microsoft Learn, Edge for Business DLP).

Edge policies allow or block, without redaction.

Edge policies for unmanaged AI apps offer allow or block, not redaction, and Microsoft notes encoded content from apps like ChatGPT can impact enforcement (Microsoft Learn, Edge for Business DLP).

Copilot prompt file uploads are not DLP-scanned.

Files uploaded directly into Microsoft 365 Copilot prompts are not scanned by DLP, and prompt SIT blocking is still in preview (Microsoft Learn, Copilot DLP). Third-party AI coverage is billed through pay-as-you-go meters whose rates are partly unpublished, which makes cost hard to forecast (Azure Purview pricing page). Frequent renames and the 2026 move from DSPM for AI to unified DSPM, with classic retirement in March 2027, add change management for admins (MC1481315).

Best fit

Organizations standardized on Microsoft 365 E5 or the Purview Suite, with Intune-managed Windows devices and an existing sensitivity label taxonomy, whose main AI exposure is Microsoft 365 Copilot, Copilot Studio agents and Foundry apps. It also suits compliance teams that want AI regulation assessments (EU AI Act, ISO 42001, NIST AI RMF) in the same console as DLP and eDiscovery.

Poor fit

Mixed-browser or Mac-heavy fleets, unmanaged or BYOD devices, and teams that need redaction or coaching inside third-party AI tools rather than allow or block. Organizations without E5 licensing or label maturity may get faster value from browser-layer tools such as Harmonic, Prompt Security or Island, and teams focused on MCP, agent runtime or red teaming of homegrown LLM apps should look at Lakera or Prompt Security.

Pricing transparency

Published on microsoft.com: Microsoft 365 E5 $60.00 per user per month (with Teams, annual), $51.45 without Teams; Microsoft Purview Suite $12.00 per user per month (annual, requires M365 E3 or O365 E3 plus EMS E3). Third-party AI app coverage uses pay-as-you-go meters billed to an Azure subscription; the Azure pricing page states audit for non-Microsoft AI apps at $15 per 1M records and Data Lifecycle Management at $6 per 1M text messages, with other AI meters (in-transit protection per 10K requests, Insider Risk per DSPU, Communication Compliance per 1K records) shown without figures.

Alternatives

See how it compares in our Best AI DLP Tools 2026 ranking. The poor-fit section above names alternatives for buyers it does not suit, and our methodology explains how every product is scored.

What We Would Test in the Lab

If Microsoft Purview DSPM for AI grants lab access, we would run the following scenarios and upgrade this review to Lab Tested.

Sensitive data detection.

The standard sensitive-data set (PII, PHI, payment data, secrets, source code).

Prompt injection.

Indirect prompt-injection scenarios from public OWASP and Lakera-style test sets.

Policy enforcement.

Block, warn, redact and allow behaviors against a configured policy.

Audit logging.

What is logged, what is not, and retention behavior.

SSO integration.

Microsoft Entra ID and Okta where supported.

Deployment.

Time from access to enforced policy in a representative environment.

Latency.

Added latency on standard prompt sizes.

Feature coverage at a glance

Based on public documentation. “Not documented” means we could not find it in public material, not that the product lacks it.

CapabilityMicrosoft Purview DSPM for AI
Inline prompt DLPYes. Copilot location blocks prompts with SITs (preview); Edge for Business and Endpoint DLP block or warn on pasting sensitive data into third-party AI sites.
Shadow AI discoveryYes. DSPM AI observability and activity explorer list AI apps and agents used in the last 30 days, using browser activity and the Defender for Cloud Apps generative AI catalog.
Blocking unapproved AI toolsPartial. Purview DLP blocks sharing of sensitive content to AI apps; blocking access to an AI site outright relies on other Microsoft products such as Entra Internet Access or Defender for Cloud Apps.
File upload inspectionYes. Edge for Business and Endpoint DLP inspect file uploads (Edge up to 3 MB); files uploaded directly into Microsoft 365 Copilot prompts are not scanned.
Secure AI workspace (sanctioned chat)Not documented. Purview protects Microsoft 365 Copilot Chat but does not itself provide a sanctioned chat front end to third-party models.
Built-in AI vendor risk scoringPartial. AI observability shows how many apps and agents are high risk; vendor-level risk ratings come from the Defender for Cloud Apps catalog, a separate product.
Microsoft 365 connected AI app review and controlPartial. Covers Entra-registered AI apps and Agent 365 agents; OAuth consent review and revocation sit in Entra and Defender for Cloud Apps rather than Purview.
Google Workspace connected AI app review and controlNot documented. No Google Workspace connected-app governance documented in Purview AI pages.
Governance scorecard, attestation and board reportingYes. Compliance Manager AI regulation assessments with exportable reports, DSPM data security objectives, posture trends and reports.
AI agent and MCP securityPartial. DLP, Insider Risk, audit and retention extend to Copilot Studio, Foundry and Agent 365 agents; MCP-specific controls are not documented.
Prompt injection defensePartial. Insider Risk 'Risky AI usage' detects prompt injection attempts and Copilot DLP can exclude external email from grounding; real-time prompt injection blocking is in other Microsoft products (Entra AI Prompt Shield preview, Foundry).
AI red teamingNot documented. No red teaming in Purview; Microsoft offers AI evaluation tooling in Foundry, which is outside this review's scope.
SIEM integrationYes. AI interactions land in the unified audit log and activity explorer, with alerts and investigations in Defender XDR.

Recent changes

  • 2025-04: Purview announced inline browser protection in Edge for Business and network data security through SASE integrations.
  • 2025-05: Purview SDK public preview, DLP for Microsoft 365 Copilot agents and Purview integration with Azure AI Foundry announced.
  • 2025-11: Ignite 2025 brought DLP, Insider Risk, audit and compliance for agents, enhanced DSPM, and Microsoft Agent 365 in preview.
  • 2026-04: Security Dashboard for AI reached general availability.
  • 2026-05: Unified DSPM became the current experience; prior versions renamed DSPM (classic) and DSPM for AI (classic).
  • 2026-10: Copilot DLP documentation extended to Microsoft 365 Copilot Cowork and added external email exclusion (preview).

Sources

Deployment and coverage

Tenant-native in the Purview portal (no separate vendor). Copilot protections need only Purview policies. Third-party AI site coverage needs the Purview browser extension and device onboarding, or Edge for Business inline protection, which for unmanaged AI apps requires Intune-managed Windows 10/11 devices and auto-created Edge and Intune policies; managed-app protection needs Conditional Access app control. DSPM reports need about a day to populate; default oversharing assessment results take about 4 days the first time.

Microsoft 365 Copilot, Copilot Chat, Cowork, Security Copilot, Copilot in Fabric, Copilot Studio, Facilitator, Teams channel agent; enterprise AI apps (Microsoft Foundry, Entra-registered AI apps, ChatGPT Enterprise, Claude Enterprise); other AI apps detected through browser activity (e.g. ChatGPT consumer, Gemini, DeepSeek, Perplexity, Grok, Meta AI); Microsoft Agent 365 agents; custom apps via the Purview SDK; network traffic to AI apps via SSE/SASE partners. MCP traffic coverage is not documented.

FAQ

Is Microsoft Purview DSPM for AI being retired?

The standalone DSPM for AI experience is now ‘DSPM for AI (classic)’. Its features moved into the unified DSPM experience, and Microsoft message MC1481315 says the classic experiences retire between mid and late March 2027.

Yes, on onboarded Windows devices through Endpoint DLP (block or warn with override) and in Edge for Business on Intune-managed devices, which can block text and file uploads to supported consumer AI apps.

DSPM and the Copilot protections are part of Microsoft 365 E5 ($60 per user per month list) or the Purview Suite add-on ($12 per user per month); third-party AI apps and network signals add pay-as-you-go Azure meters.

Insider Risk Management’s Risky AI usage template detects prompt injection attempts for risk scoring, and Copilot DLP can exclude external email from grounding; it is not a real-time prompt firewall for custom apps.