🧠 AB-650 Course · Module 6 of 9
Purview Data Protection & AI Governance
AB-650 Exam Alignment
AB-650
Domain 2 — Implement Microsoft Purview data protection and governance: implement DLP policies for Exchange, SharePoint, OneDrive, Teams, endpoints, and Copilot; implement information protection using SITs, sensitivity labels, and label policies; implement data lifecycle management with retention labels and policies; review DLP alerts for M365 and AI services; implement Purview DSPM for AI to monitor and secure AI activity.
- Know the four sensitivity label scopes: files & emails, meetings, groups & sites, schematised data assets
- Understand the difference between DLP policy tips (informational, shown to user) vs DLP alerts (sent to admins/security team)
- Know that Purview DSPM for AI is specifically designed to monitor AI interactions — including Copilot prompts and responses that may contain sensitive data
- Understand retention label vs retention policy: labels apply to individual items; policies apply to entire locations (all Exchange mailboxes, all SharePoint sites)
Exam Tip: AB-650 extends DLP to AI services — specifically, Copilot interactions can be monitored and protected by DLP policies. Know that DLP for Copilot requires Microsoft Purview to be configured to monitor “Microsoft 365 Copilot” as a location in the DLP policy — this is distinct from the Exchange, SharePoint, and Teams locations.
Microsoft Purview is the compliance and information protection platform for Microsoft 365. In the AB-650 era, Purview extends to AI services — Copilot prompts, responses, and agent interactions can now contain sensitive data that must be governed by the same DLP, retention, and information protection policies applied to traditional M365 workloads. This module covers the complete Purview data governance stack with a focus on the new AI-specific capabilities.
🔒 Data Loss Prevention (DLP)
DLP policies prevent the sharing of sensitive information — credit card numbers, health records, national IDs — across M365 workloads. In AB-650, DLP extends to Copilot interactions as a dedicated location.
| DLP Location | What It Monitors | Possible Actions |
|---|---|---|
| Exchange Online | Email content including attachments | Block send, encrypt, notify sender, generate alert |
| SharePoint Online | Files stored in SharePoint sites | Block sharing, restrict access, apply label, alert |
| OneDrive | Files in OneDrive for Business | Block sharing, restrict access, apply label, alert |
| Microsoft Teams | Chat messages and channel messages (content, not files separately) | Block message, notify user, generate alert |
| Endpoints (MDE) | Files on Windows 10/11 devices enrolled in Microsoft Defender for Endpoint | Block copy to USB, block upload to cloud, block print, block clipboard operations |
| Microsoft 365 Copilot 🔥 | Prompts submitted to Copilot and Copilot responses that contain or reference sensitive data | Block Copilot response, display policy tip in Copilot interface, generate compliance alert |
🔥 New in AB-650: DLP for Microsoft 365 Copilot
DLP policies can now target Microsoft 365 Copilot as a dedicated location. When sensitive information (e.g. credit card numbers, NHS numbers, patient data) appears in a Copilot prompt or Copilot surfaces it in a response from SharePoint or Exchange, DLP can: (1) block the Copilot response from being shown to the user; (2) display a policy tip within Copilot Chat informing the user why the response was blocked; (3) generate a DLP alert in the Purview compliance portal for the security team to review. This is configured in the Microsoft Purview compliance portal → Data loss prevention → Policies → Add location: Microsoft 365 Copilot.
🏷️ Sensitivity Labels & SITs
| Concept | Description | AB-650 Key Points |
|---|---|---|
| Sensitive Information Types (SITs) | Pattern-based classifiers that detect specific data types (credit cards, SSNs, NHS numbers, passport numbers). Used as conditions in DLP policies and sensitivity label auto-labeling. | There are 200+ built-in SITs. Custom SITs can be created using regex patterns and keyword lists. Know that confidence level (Low/Medium/High) affects match accuracy vs false positive rate. |
| Sensitivity Labels | Classify and optionally protect content. A label like “Confidential” can apply encryption, add visual markings (header, footer, watermark), and restrict sharing. | Labels can be applied manually by users or automatically by auto-labeling policies (client-side or service-side). Labels persist with the content — moving a file doesn’t remove the label. |
| Label Policies | Control which users see which sensitivity labels and define default labels. Can require users to apply a label before saving or sending. | Publish labels to specific users/groups via label policies. A label must be published before it appears for users. Mandatory labelling requires a label to be applied — users cannot save without one. |
| Auto-labeling policies | Automatically apply sensitivity labels to content in Exchange, SharePoint, and OneDrive based on SIT matches — without user intervention. | Service-side auto-labeling can scan all existing content in SharePoint at rest. Client-side auto-labeling shows a recommendation or applies automatically when a user opens a file. |
📅 Data Lifecycle Management — Retention
| Feature | Scope | Key Behaviour |
|---|---|---|
| Retention policy | Entire location — all Exchange mailboxes, all SharePoint sites, all OneDrive accounts, or Teams chats | Retains all content in the specified location for the configured period. Cannot be overridden by users — content is preserved even if deleted. Does not give individual items different retention periods. |
| Retention labels | Individual items — specific emails, documents, Teams messages | Applied manually or via auto-apply policies. Different items can have different retention periods. Supports record management — items labelled as records cannot be modified or deleted. |
| Copilot interaction history | Copilot prompts and responses stored in Exchange Online (hidden mailbox folder) | Can be retained and discovered via eDiscovery and retention policies targeting Exchange. AB-650 tests how to configure retention for Copilot interaction history. |
📊 Purview DSPM for AI NEW IN AB-650
Data Security Posture Management for AI (DSPM for AI) is a Microsoft Purview feature that provides visibility and control over how AI services interact with organisational data. It is the primary tool for monitoring Copilot and agent data interactions from a compliance perspective.
🔥 Purview DSPM for AI — What It Does
- Provides a dashboard of AI-related data risks: sensitive data referenced in Copilot prompts, sensitive data surfaced in Copilot responses, data accessed by agents
- Shows which sensitivity labels are most frequently accessed by Copilot users — helps identify over-privileged data access
- Integrates with Insider Risk Management to flag unusual AI data access patterns
- Generates DLP alert reports specifically for AI service interactions
- Located: Microsoft Purview compliance portal → Information protection → DSPM for AI
- Requires Microsoft Purview Information Protection licence (included in M365 E5 Compliance or as an add-on)
Microsoft Graph PowerShell & Compliance — DLP & Sensitivity Labels
# Connect to Security & Compliance PowerShell Connect-IPPSSession -UserPrincipalName admin@contoso.com # List all DLP policies and their workload locations Get-DlpCompliancePolicy | Select-Object Name,Mode,@{N='Locations';E={$_.Workload}} | Format-Table -AutoSize # List all sensitivity labels in the tenant Get-Label | Select-Object DisplayName,Priority,IsActive,ContentType | Sort-Object Priority | Format-Table -AutoSize # List label policies (which labels are published to which users) Get-LabelPolicy | Select-Object Name,Labels,@{N='Users';E={$_.ExchangeLocation}} | Format-Table -AutoSize # Get DLP alerts summary for the last 30 days Get-DlpDetectionsReport -StartDate (Get-Date).AddDays(-30) | Group-Object PolicyName | Select-Object Name,Count | Sort-Object Count -Descending | Format-Table -AutoSize # Review retention policies Get-RetentionCompliancePolicy | Select-Object Name,Enabled,Mode | Format-Table -AutoSize
💡 Best Practices
- Start DLP policies in test/audit mode before switching to enforce mode — review DLP reports for 2–4 weeks to identify false positives before blocking content
- Use a small number of clearly-named sensitivity labels (4–6 labels max) — too many labels confuse users and reduce adoption. Start with Personal, General, Confidential, Highly Confidential.
- Enable auto-labeling for high-confidence SIT matches (e.g. High confidence credit card number) — this catches sensitive data that users forget to label manually
- Include Microsoft 365 Copilot as a DLP location in all policies that cover financial, health, or PII data — Copilot can surface this data in responses if not explicitly protected
- Review the Purview DSPM for AI dashboard monthly as part of the Copilot security review — look for unexpected sensitivity label categories being accessed frequently by Copilot users
- Configure retention for Copilot interaction history via Exchange retention policies — Copilot prompts and responses are stored in Exchange and subject to eDiscovery
🎓 Interview Q&A
Q: What is the difference between a DLP policy and a retention policy in Microsoft Purview?
A DLP policy detects and acts on sensitive information at the point of sharing or use — it prevents data leakage in real-time by blocking sends, applying encryption, or notifying users when sensitive data is detected. It does not affect how long data is kept. A retention policy governs how long content is kept and what happens when the retention period expires — it ensures data is preserved for the required period (regulatory compliance, litigation hold) or deleted after it’s no longer needed. Both can apply to the same data: a DLP policy might prevent sharing a document containing credit card data, while a retention policy ensures that same document is kept for 7 years to meet financial regulations.
Q: What is Purview DSPM for AI and what problem does it solve?
Data Security Posture Management for AI (DSPM for AI) is a Microsoft Purview feature that provides visibility into how AI services (Microsoft 365 Copilot, Agent 365) interact with organisational data. The problem it solves: Copilot users can ask the AI to summarise or retrieve any document they have access to — including documents with sensitivity labels like Confidential or Highly Confidential. Without DSPM for AI, administrators have no visibility into which sensitive data Copilot is accessing and surfacing in responses. DSPM for AI shows a dashboard of AI data access patterns, highlights which sensitivity label categories are most frequently accessed, identifies users whose Copilot interactions involve high volumes of sensitive data, and integrates with DLP alerts specific to AI service interactions.
Q: A user wants to apply a sensitivity label to an email in Outlook, but the label is not visible. What is the most likely reason?
The most likely reason is that the sensitivity label has not been included in a label policy published to that user. In Microsoft Purview, creating a label does not make it visible to users — the label must be published via a label policy that targets specific users or groups. To resolve: go to Microsoft Purview compliance portal → Information protection → Label policies, create or edit a policy to include the label, and ensure the user is in the scope of the policy. Note that label policy changes can take up to 24 hours to propagate to client applications.
🎯 AB-650 Mock Test
Module 6 — Purview Data Protection & AI Governance
Question 1 of 5
A healthcare organisation wants to prevent Copilot from surfacing patient health record data (NHS numbers) in its responses. Which configuration achieves this?
✅ Correct: C. A DLP policy targeting the Microsoft 365 Copilot location with the NHS Number sensitive information type as a condition will detect when Copilot attempts to include NHS numbers in its responses and block the response. This is the specific AB-650 capability for preventing sensitive data from being surfaced by Copilot. Option A (sensitivity label) classifies content but doesn’t block Copilot from accessing it. Option B (Restricted Content Discovery) prevents Copilot from discovering the site but doesn’t prevent it from accessing data the user opens directly. Option D blocks Copilot access entirely rather than selectively protecting specific data types.
Question 2 of 5
A compliance officer wants all SharePoint documents containing credit card numbers to be automatically labelled “Confidential — Financial” without requiring users to apply labels manually. Which Purview feature achieves this?
✅ Correct: A. Service-side auto-labeling policies in Microsoft Purview scan content in SharePoint, Exchange, and OneDrive and automatically apply sensitivity labels based on SIT matches — without any user interaction. They can scan existing content at rest (not just new content). Configure in Purview compliance portal → Information protection → Auto-labeling. DLP policies (B) take actions (block, alert) but don’t apply sensitivity labels. Label policies (C) publish labels to users but require manual application. Retention policies (D) manage data lifecycle, not classification.
Question 3 of 5
Which Microsoft Purview feature provides a dashboard specifically showing which sensitivity label categories are most frequently accessed by Microsoft 365 Copilot, helping administrators identify AI-related data risks?
✅ Correct: D. Purview DSPM for AI (Data Security Posture Management for AI) is the specific feature that provides visibility into how AI services access and surface sensitive data. It shows which sensitivity label categories Copilot users are accessing most frequently, highlights over-privileged data access patterns, and integrates with compliance alerts. The Copilot Dashboard (B) shows usage and adoption — not data security posture. DLP reports (C) show policy matches but not AI-specific access patterns. Secure Score (A) tracks overall security posture, not AI-specific data governance.
Question 4 of 5
An administrator creates a new sensitivity label “Project-Alpha-Confidential” and applies it to a test document. Users report they cannot see the label in Microsoft Office apps to apply it to their own documents. What step was missed?
✅ Correct: B. Creating a sensitivity label in Microsoft Purview does not make it visible to users. The label must be included in a label policy and published to users or groups. Create or edit a label policy in Purview → Information protection → Label policies, add the new label, and target the appropriate users. Changes take up to 24 hours to appear in Office apps. Encryption (A) is optional and not required for label visibility. Sensitivity labels are available with Microsoft 365 E3+ (C — not E5-only). There is no separate “activation” step in the Admin Center (D).
Question 5 of 5
What is the correct way to ensure that Copilot interaction history (prompts and responses) is retained for 3 years for compliance purposes in a Microsoft 365 tenant?
✅ Correct: C. Microsoft 365 Copilot interaction history (prompts and responses) is stored in a hidden folder within the user’s Exchange Online mailbox — similar to how Teams chat history is stored. To retain this data, configure an Exchange retention policy targeting user mailboxes. This data is also discoverable via Microsoft Purview eDiscovery. SharePoint (A) stores documents and files, not Copilot interaction history. Teams retention (B) covers Teams chat messages, not Copilot Chat. DSPM for AI (D) provides visibility and alerts but is not a retention configuration tool.
🔒
This module is lockedComplete Module 5 and pass its mock test to unlock this module.