24.3 Configure DSPM for AI Policies

Configure DSPM for AI Policies

Quick Answer: Open Recommendations (or Get started) in DSPM for AI (classic), activate the one-click policies that match your AI surfaces, then use the Policies page only as a status board. Edit collection policies, DLP, Insider Risk Management, Communication Compliance, and sensitivity labels in their own solutions. For Fabric and Security Copilot prompt capture, take Secure interactions in Microsoft Copilot experiences. For enterprise apps, take Secure interactions from enterprise apps.

Configuring DSPM for AI policies is not the same as authoring a blank DLP policy. Microsoft ships preconfigured one-click policies that you activate from personalized recommendations. After they exist, they are ordinary Purview policies with DSPM-style names. SC-401 expects you to know which recommendation creates which policy, which solution owns it, what mode it starts in, and what you still have to configure by hand (SASE/SSE, content capture, label selection, scoping during a pilot).

Allow at least 24 hours after creation or default-setting changes before Reports reflect the policy. If you use administrative units, only an unrestricted administrator can create the one-click policies that apply to all users.

How one-click policies fit the rest of Purview

DSPM for AI is the front door. The policies it creates land in:

• Collection policies — ingest and optionally store AI prompts and responses, and detect sensitive information on network or Edge paths
• Data Loss Prevention — detect or block sensitive content going to AI sites or being processed by Copilot
• Insider Risk Management — AI site visits and risky AI usage scoring
• Communication Compliance — unethical behavior in AI apps
• Information Protection — default sensitivity labels and label policies, skipped if you already have labels

The Policies page lists those objects and their status. To change scope, classifiers, or actions, open the owning solution. Example: matches for DSPM for AI - Unethical behavior in Copilot (Microsoft also documents the name DSPM for AI - Unethical behavior in AI apps) are reviewed in Communication Compliance, not inside a DSPM-only queue.

Older retention policies that still use the combined location Teams chats and Copilot interactions do not appear on this Policies page. Create separate retention policies for Microsoft Copilot Experiences if you want those retention policies included there.

If you delete a one-click policy, status becomes PendingDeletion. The recommendation card can continue to show the policy as created until deletion finishes. Policies created while the feature was in preview as Microsoft Purview AI Hub keep the Microsoft AI Hub - name prefix; Microsoft does not rename them automatically.

Recommendations that create policies

Work Recommendations (Overview or the Recommendations pane). Completed or dismissed recommendations leave Overview. View all recommendations is the full list. Typical cards include:

• Protect your data from potential oversharing risks — review the default data risk assessment rather than creating a policy
• Protect your data with sensitivity labels — creates default labels and policies only if you do not already have them
• Protect items with sensitivity labels from Microsoft 365 Copilot and agent processing — you select one or more labels; the DLP policy then blocks Copilot and agents from summarizing those labeled items
• Detect risky interactions in AI apps — Insider Risk Management risky AI usage
• Discover and govern interactions with ChatGPT Enterprise AI — register ChatGPT Enterprise workspaces
• Get guided assistance to AI regulations — Compliance Manager regulatory templates (control mapping, not a DLP policy)
• Secure interactions for Microsoft Copilot experiences — collection policy for Fabric and Security Copilot
• Detect sensitive info shared with AI via network — collection policy that needs SASE/SSE
• Secure interactions from enterprise apps — collection policy for Entra-registered apps, ChatGPT Enterprise connector, and Microsoft Foundry apps
• Secure data in Azure AI apps and agents — setup instructions to capture prompts and responses for apps that use Azure AI subscriptions
• Fortify your data security — Adaptive Protection and Edge DLP block policies
• Extend your insights for data discovery — the Get started discovery set (DLP audit-mode, IRM visit detection, Edge SIT detection)

Default policies for data discovery

Collection policies and content capture

Collection policies are the SC-401 center of this skill. They filter which events Purview ingests. DSPM one-click actions create collection policies for you; you can later open them under Data Loss Prevention > Classifiers > Collection policies or Information Protection > Classifiers > Collection policies and edit them like any other collection policy.

Content capture for AI interactions stores detected prompts and responses from generative AI data sources added to the policy. It does not include content in files shared with generative AI. Microsoft documents that capture applies to Copilot experiences, Enterprise AI, unmanaged cloud apps categorized as generative AI, and the all unmanaged AI apps adaptive app scope. Without capture, detection is limited to sensitive information only.

To capture AI content you must set the Content contains classifiers condition to All. The one-click network policy does not select content capture when it is auto-created. If the exam says you see SITs but not full prompts from a network or Edge collection policy, edit the policy and enable capture (and set classifiers to All). The Edge discovery policy is explicitly audit mode and does not capture content until you change it.

DSPM for AI - Detect sensitive info shared with AI via network has a second manual step: add one or more SASE or SSE integrations in Data Loss Prevention settings. Detection depends on the network partner implementation. Creating the named policy without the integration produces an empty signal path.

Microsoft 365 Copilot prompt capture is not this Fabric/Security Copilot collection policy. Copilot in Microsoft 365 continues to flow through audit. Know both pipelines so you do not disable the wrong object when a privacy team asks you to stop storing prompts.

Default policies that protect data in generative AI

Default policies that use Adaptive Protection turn Adaptive Protection on if it is not already on, using default risk levels for all users and groups. That is a side effect you must expect: taking Fortify your data security is not “just a test DLP rule.”

After creation, scope the policies to a pilot group, add or remove classifiers, or move DLP from test/audit to enforce — in the DLP or collection-policy UI. DSPM will not magically keep a second copy of the settings.

Scenario: enterprise ChatGPT versus consumer ChatGPT

A bank allows ChatGPT Enterprise and wants prompts available to eDiscovery. Take Secure interactions from enterprise apps, which creates DSPM for AI - Capture interactions for enterprise AI apps, and complete ChatGPT Enterprise workspace registration if that recommendation is shown. Consumer ChatGPT in a browser is Other AI apps / third-party sites: you need devices, the browser extension, Extend your insights, and possibly endpoint or Edge DLP. Do not expect the enterprise collection policy to see unmanaged consumer sessions.

Exam traps

• One-click is not “set and forget” for the network policy. SASE/SSE integration is manual.
• Content capture is a collection-policy checkbox plus classifiers = All, not a separate DSPM product.
• Protect sensitive data from Copilot processing is a DLP policy driven by labels you pick, not a magical block of all Copilot.
• Audit mode and test mode are not the same as block. The discovery DLP policy starts in audit; the Adaptive Protection paste policy starts in test.
• Do not edit from memory inside DSPM. Follow the Policies page into DLP, IRM, Communication Compliance, or collection policies.

Official references: One-click policies, Collection policies.