SC-401 protecting data used by AI services
Protect data used by AI services is the newest part of SC-401 and a sub-area of Manage risks, alerts, and activities. The idea is simple: AI assistants can only surface what users can already reach, so the controls are the ones you already know, plus Data Security Posture Management (DSPM) for AI to see what is happening and to deploy policies quickly. The exam asks about prerequisites, roles, policies and monitoring, and about which existing Purview and Microsoft 365 control limits AI exposure.
Why this is on the exam
Microsoft’s audience profile for the role now says it includes protecting data used by AI services. If a tenant has overshared files, an AI assistant can quote them to someone who technically has access but should never have found them. The administrator’s job is to find that exposure and tighten it.
DSPM for AI
DSPM for AI is a Purview solution that gives visibility into how AI services touch sensitive data and offers recommended policies. Think of four tasks, which match the skills outline:
| Task | What it involves |
|---|---|
| Prerequisites | Auditing turned on; devices onboarded where endpoint signals are needed; the browser extension for monitoring other AI sites where required |
| Roles and permissions | Assigning the right Purview or Entra roles so that the right people can configure DSPM for AI and others can only view reports |
| Policies | Creating recommended policies, for example to detect sensitive data in prompts or to stop sensitive content being processed |
| Monitoring | Reviewing AI activity, sensitive interactions and risk in the DSPM for AI reports and in Activity explorer |
Because DSPM for AI builds on audit data, a tenant that has not turned on auditing will see little. Questions that ask why the dashboard is empty frequently have that answer.
Existing controls that protect AI use
- Sensitivity labels: encryption on a label restricts who can use the content, and AI services respect those usage rights. A label that does not allow extraction by a user stops the assistant from returning that content to them.
- DLP: DLP policies can cover AI interactions, for example by identifying sensitive info types in prompts or by preventing labelled content from being processed.
- Insider Risk Management: risky AI usage can be treated as an indicator and feed Adaptive Protection.
- Audit and eDiscovery: AI interactions are recorded and can be searched, so compliance teams can investigate and retain them like other communications.
- Reducing oversharing: reviewing permissions on sites and files, and applying labels to sensitive content, reduces what assistants can reach in the first place.
Common pitfalls
Treating AI protection as separate from the rest of Purview is the main mistake. The strongest answers on the exam extend existing controls rather than invent new ones. Another mistake is assuming protection works without prerequisites: no auditing, no endpoint onboarding and no extension means no signal.
Three example questions
Question 1. You open DSPM for AI in a new tenant and see almost no data. What should you check first?
- A. That a retention policy exists
- B. That a label policy is published
- C. That an IRM policy uses the Data leaks template
- D. That Microsoft Purview auditing is turned on
Show answer
Answer: D
DSPM for AI depends on audit data, so auditing must be enabled. A retention policy, a sensitivity label policy and an IRM template are unrelated to the missing activity data.
Want more questions like this? Full SC-401 practice tests →
Question 2. Executives worry that an AI assistant could summarise HR files for employees who technically have access but should not. Which approach reduces exposure most directly?
- A. Disable auditing
- B. Label sensitive files with restrictive protection and fix oversharing
- C. Extend audit retention to ten years
- D. Place all HR users on eDiscovery hold
Show answer
Answer: B
Applying sensitivity labels with access restrictions to sensitive files, and reviewing overshared permissions, limits what the assistant can return. Disabling audit removes visibility, extending audit retention preserves logs but does not limit access, and an eDiscovery hold preserves data rather than restricting it.
Want more questions like this? Full SC-401 practice tests →
Question 3. A security administrator needs to review AI risk reports but must not be able to change DSPM for AI policies. What should you do?
- A. Assign a read-only role for DSPM for AI
- B. Assign the Global Administrator role
- C. Disable auditing
- D. Remove the person from all role groups and email the reports
Show answer
Answer: A
Assign a view-only role for DSPM for AI so the person can read reports without configuration rights. Giving a full administrator role breaks least privilege, and disabling the portal or auditing prevents the review.
Want more questions like this? Full SC-401 practice tests →
What to practise
Open DSPM for AI in a test tenant, read the prerequisites list, apply a recommended policy and review the resulting activity. Pair this with the labels guide and the DLP guide.