2.4.5. Data Security Posture Management for AI Activity
💡 First Principle: You can't secure AI usage you can't see. DSPM's job is to observe where sensitive data meets AI — Copilot, agents and third-party AI apps — and turn what it finds into policies, so posture improves continuously instead of in one-off audits.
Name check. The current product is Microsoft Purview Data Security Posture Management (DSPM), at Purview portal > Solutions > DSPM, generally available since May 2026. The earlier experiences are now labeled DSPM (classic) and DSPM for AI (classic), and most new features ship only in the new version. The skills outline's "Monitor and secure AI activity by using DSPM" refers to this experience.
The pages that matter for AI:
| Page | Use it to... |
|---|---|
| Objectives | Work toward goals such as Prevent data exposure in Microsoft 365 Copilot and Microsoft Copilot interactions or Prevent oversharing, each with a remediation plan and one-click policies |
| AI observability | See the AI apps and agents active in the last 30 days — including Agent 365 agents — with risk and sensitive-interaction counts |
| Activity explorer (AI activities tab) | Review prompts and responses that contained sensitive information, and DLP matches |
| Data risk assessments | Find oversharing — the default assessment runs weekly on the top 100 SharePoint sites by usage, with first results after about four days; custom assessments can target specific users or sites |
| Remediation actions | Find the automatic and one-click policies (most came over from DSPM for AI) |
One-click policies from DSPM create starter policies in other Purview solutions — for example, a DLP policy that detects sensitive information pasted or uploaded to AI sites in the browser (audit only), an Insider Risk policy that detects visits to AI sites, and a Communication Compliance policy for unethical behavior in Copilot. Data risk assessment remediations include applying a sensitivity label, notifying the site owner and removing sharing links; item-level scanning covers SharePoint (up to 10 sites), not OneDrive.
Prerequisites for third-party AI apps (ChatGPT or Gemini in a browser): Purview Audit turned on (the default for new tenants), devices onboarded to Purview, and the Purview browser extension. Without device onboarding, DSPM can't see — or warn about — sensitive data pasted into a consumer AI site.
Permissions: the Entra Compliance Administrator role or the Purview Compliance Administrator role group to set it up. The AI Administrator can view DSPM but not configure it. After first-time setup, allow about a day before data appears.
⚠️ Exam Trap: "Use DSPM for AI to…" describes the classic experience; the current answer is DSPM. And DSPM's visibility into consumer AI websites depends on device onboarding and the browser extension — not on a Copilot license.
Reflection Question: A CISO asks, "Which AI tools are our people pasting customer data into, and how do we stop it?" Which DSPM pages answer the first half, and which prerequisites must be in place before you can enforce the second?