Optional Lab Only

This optional lab requires a Microsoft 365 tenant that you are authorized to administer, the required Microsoft Purview licensing, and the permissions listed below.

The Purview DSPM experience is still evolving. Validate the current portal experience before delivering the lab.

Please read through to learn the steps that are required. We apologize for the delay in this lab.

===

Identify AI Data Risks with Microsoft Purview

Your organization deployed Microsoft 365 Copilot six months ago. No one has reviewed what data Copilot can access on behalf of users since deployment. A security review request asks you to identify which SharePoint sites are overexposed and what risk signals Purview is surfacing about Copilot data access — before the next compliance audit.

As a Cloud and AI Security Engineer, you are not an information protection worker. You will not be configuring Purview policy, applying sensitivity labels, or changing SharePoint permissions. Your job is to navigate the Data Security Posture Management dashboard, understand what each finding means as a security risk, and determine which remediation actions to recommend and to whom.

In this lab, you will:

  • Navigate the Microsoft Purview DSPM for AI (classic) dashboard
  • Identify a SharePoint site with oversharing conditions that make it accessible to Copilot
  • Identify unlabeled sensitive data that Copilot can surface without restriction
  • Review Copilot interaction signals to understand what data Copilot has already accessed
  • Evaluate recommended remediations and identify who owns each action

This exercise should take approximately 45 minutes to complete.

Prerequisites and Setup

Important: This lab requires a pre-configured SharePoint site named sc500-ai-datastore with oversharing permissions and sample sensitive documents.

Setup Required Before Starting This Lab

Option 1 - Automated Setup (Recommended):

  1. Download the setup script from Allfiles\Lab-3A\lab-3a-setup.ps1
  2. Run in PowerShell 7: 
    .\lab-3a-setup.ps1 -TenantUrl "https://YOUR-TENANT.sharepoint.com"
  3. Wait 2-3 minutes for completion

Option 2 - Manual Setup:

  • Follow the detailed instructions in Allfiles\Lab-3A\README.md

Option 3 - Existing setup:

  • If the site is already configured, verify it by navigating to https://YOUR-TENANT.sharepoint.com/sites/sc500-ai-datastore

Timing Note: Purview DSPM signals can take 24-48 hours to fully populate after site creation. In a lab environment, you can proceed immediately to explore the dashboard structure, though some risk signals may still be processing.

Open the Purview Data Security Posture Management dashboard

  1. Sign in to the Microsoft Purview compliance portal https://purview.microsoft.com using your User-1 credentials.

  2. Select Solutions > DSPM for AI (classic).

    Note: Look for DSPM for AI (classic) in the Solutions list — not DSPM (the newer version, which requires a first-time configuration wizard) and not Data Security Posture Management (classic). DSPM for AI (classic) provides the focused AI data risk views this lab uses and is available immediately without additional setup.

  3. The Overview page opens. Review the Get started and Recommendations sections — you should see risk indicators related to SharePoint oversharing, unlabeled sensitive data, and Copilot interaction signals.

    Note: This is a lab environment with a pre-seeded SharePoint site, so signal counts will be small. In a production tenant these numbers can reach into the thousands. The objective is to understand what each signal type means and how to evaluate it, not to process volume.

Identify the SharePoint oversharing risk

SharePoint oversharing is the most common AI data risk in organizations that have deployed Copilot. When a site is shared with Everyone or Everyone except external users, Copilot can surface that content for any user who queries it — including documents that user would never think to look for directly. DSPM surfaces these sites so the risk can be identified and escalated.

  1. In the Recommendations section on the Overview page, select Protect your data from potential oversharing risks, or in the left navigation select Data assessments.

  2. Locate sc500-ai-datastore in the assessment results.

  3. Select sc500-ai-datastore to open the site-level exposure details.

  4. Review the Access configuration section. Confirm the site is shared with Everyone except external users at the site level — this is the oversharing condition DSPM flagged.

  5. Review the Sensitive content section. Note the document types listed — look for any documents flagged as containing personal information, financial data, or other sensitive categories.

  6. Review the Recommended actions for this site. You should see actions similar to the following:

    Recommended action Who owns it
    Restrict site permissions to specific groups SharePoint administrator or site owner
    Apply sensitivity labels to flagged documents Information protection team
    Initiate an access review for the site Security engineer escalates — site owner executes

    Note: As a security engineer, you identify the finding and produce a clear recommendation. You do not change SharePoint permissions directly — that requires site owner involvement and organizational authority. Your deliverable is the finding and the remediation recommendation.

Review unlabeled sensitive data

Sensitivity labels tell Copilot how to treat a document — whether it can summarize, reference, or share its contents. Unlabeled documents give Copilot no guidance, and it can use them freely. DSPM flags unlabeled sensitive content so the organization knows where labeling gaps create AI risk.

  1. Return to the Overview page and locate the Unlabeled sensitive data section, or select the equivalent item in the left navigation.

  2. Locate files associated with sc500-ai-datastore that are flagged as unlabeled sensitive content.

  3. Review the detected content type for each flagged file — note whether the content is classified as HR data, financial data, personally identifiable information, or another category.

  4. Confirm that none of the documents in sc500-ai-datastore have sensitivity labels applied. In a remediated state, each of these documents would carry a label that controls what Copilot can do with the content.

    Note: Applying sensitivity labels is not a task for this lab. Label policy configuration is a compliance engineering role, not a security engineering role. Your finding here is: unlabeled sensitive data exists in a Copilot-accessible location and is not governed by any label-based access restriction.

Review Copilot interaction signals

The Copilot interactions section shows what data Copilot has recently accessed on behalf of users. This is the evidence that over-permission translates into actual exposure — Copilot does not need a security misconfiguration to access data; it only needs the invoking user to have access.

  1. Return to the Overview page and locate the Copilot interactions section, or select the equivalent item in the left navigation.

  2. Review the recent Copilot interaction entries. Look for any interactions that reference content from sc500-ai-datastore.

  3. Note the type of data accessed in those interactions — for example, a document summarization query that pulled from the pre-seeded HR data in the site.

  4. Consider the attack path this creates: if an attacker compromises a user account that has access to sc500-ai-datastore, they can use Copilot to rapidly extract and summarize sensitive content from the site without ever browsing to the files. Copilot makes existing over-permission faster and more powerful — DSPM's job is to make that risk visible before it is exploited.

Evaluate the findings

  1. Based on your review, confirm that sc500-ai-datastore presents three distinct risk signals:

    Risk signal Root cause Security implication
    Oversharing Site shared with Everyone All users — and Copilot acting on their behalf — can access HR and financial content
    Unlabeled data No sensitivity labels applied Copilot can reference and surface document content with no policy restriction
    Copilot interaction exposure Copilot has already accessed content on behalf of users The theoretical risk has already become an actual access event

  2. For each risk signal, confirm you can identify the recommended remediation and the role that owns the action. A security engineer's output from this review is an escalation to the site owner (for permissions) and a referral to the information protection team (for labeling). The security engineer does not directly execute either action.

  3. Select Overview in the left navigation to return to the DSPM for AI (classic) main view and confirm no other high-priority signals require immediate escalation in this environment.

Summary

In this lab, you navigated the Microsoft Purview DSPM for AI (classic) dashboard and identified three AI data risk signals associated with a pre-seeded SharePoint site: a site-level oversharing condition that makes its content accessible to Copilot for all users, unlabeled sensitive documents that carry no policy-based access restriction, and evidence of actual Copilot interactions that accessed that content.

You reviewed the recommended remediations for each finding and identified that the remediation actions belong to the SharePoint administrator, the information protection team, and the site owner — not to the security engineer. As a Cloud and AI Security Engineer, your contribution is the posture visibility review: finding the risk, understanding what it means, and producing a clear recommendation for the people who own the fix.

You have successfully completed this exercise.

Clean up

No resources were created in this lab. No cleanup is required.