SC-401 and DSPM for AI: The Exam Objective SC-400 Never Had

Of everything that changed when SC-401 replaced SC-400, one addition stands out: an objective group called Protect data used by AI services. It did not exist on SC-400, and it centres on a Purview capability called Data Security Posture Management (DSPM) for AI.

It sits inside Manage risks, alerts, and activities — a domain weighted 30 to 35 percent. Candidates converting from SC-400 material almost universally arrive with nothing on it.

Why this objective exists

The problem it answers is concrete. Organisations switched on generative AI — Microsoft 365 Copilot, custom agents, third-party assistants — and those tools read organisational data at a speed and breadth no person could. A document with the wrong permissions used to be a latent risk; with an AI assistant in the tenant it becomes an immediately retrievable answer.

Traditional data security controls were built around where data sits and who opens it. AI introduced a third question: what is being fed into a model, and what came back? DSPM for AI exists to make that question answerable.

What DSPM for AI actually is

DSPM for AI is the Purview surface for discovering and governing how AI interacts with your data. Practically it gives you three things:

• Discovery — which AI applications are being used across the organisation, including ones nobody formally approved.
• Insight into interactions — visibility into AI activity and whether sensitive data is involved in it.
• Policy — the ability to act on what you find, rather than only observe it.

It does not replace sensitivity labels, DLP or Insider Risk Management. It sits on top of them and gives them an AI-shaped view. That relationship is worth holding onto, because it explains several exam objectives at once.

The six exam objectives

Microsoft lists exactly six bullets under Protect data used by AI services. Learn them as a sequence — they describe an implementation order, not an unordered list.

The first two are about the ground you stand on; the last four are DSPM for AI itself.

Objectives 1 and 2: the controls underneath

These two ask you to protect AI-touched content using capabilities you already know. In Purview that means sensitivity labels with real protection settings, DLP policies that catch sensitive content in motion, and retention where it applies. In the productivity workloads it means the permission and sharing model in SharePoint, OneDrive and Teams.

Prerequisites

Objective 3 is implement prerequisites for DSPM for AI — phrased as a task, which means knowing that prerequisites exist is not enough.

Think about it in three layers, and verify the current specifics in Microsoft's documentation before your exam, since this area changes:

• Licensing — the capability requires appropriate Purview licensing, and related capabilities such as Purview Audit (Premium) have their own licence requirements, which SC-401 tests separately.
• Onboarding and signal — DSPM for AI can only report on activity it can see. Devices and workloads need to be onboarded so that activity flows in; without that the dashboard is empty and every downstream objective is moot.
• Underlying classification — telling you sensitive data was involved requires that sensitive data be identifiable. Sensitive information types, trainable classifiers and labels have to be in place first.

Roles and permissions

Objective 4. SC-401 tests roles and permissions repeatedly — there is a similar objective for sensitivity labels, for DLP and for Insider Risk Management. Microsoft is making a consistent point: in Purview, what you can see and do is scoped by role, and least privilege is the expected answer.

For DSPM for AI, be clear on the difference between someone who can view the posture and reports, and someone who can configure policies. Those are different responsibilities and, in a well-run tenant, different people. When a question describes an analyst who needs to investigate AI activity but should not change policy, it is testing exactly that distinction.

Policies

Objective 5. DSPM for AI provides policies that let you act on discovered AI activity rather than merely observe it — including policy templates aimed at common AI risks, which is the fastest route from discovery to control.

Two connections the exam is likely to lean on:

• To DLP. Preventing sensitive content from reaching an AI service is a data loss prevention problem, and DLP is a neighbouring domain in the same exam.
• To Insider Risk Management. A user routing sensitive material through unsanctioned AI tools is an insider risk signal. Adaptive Protection — also new in SC-401 — is the mechanism that lets protection strength follow assessed risk.

SC-401 rewards seeing these as one system. The three equally weighted domains are not three separate subjects; they are three views of the same data-security problem.

Monitoring activities

Objective 6, and the one that closes the loop. Once prerequisites are met, roles assigned and policies configured, you have to keep looking.

Monitoring in DSPM for AI overlaps with tools tested elsewhere in the exam — Activity explorer for Purview activity, Purview Audit for investigation, and alert response in the Purview portal and Microsoft Defender XDR. Know which tool answers which question:

How it connects to the rest of SC-401

Do not study this objective group in isolation — the exam does not test it in isolation.

• Information protection supplies the classification and labels without which DSPM for AI cannot tell you anything meaningful.
• DLP and retention supply the enforcement that stops sensitive content reaching AI services, and govern what is kept afterwards.
• Insider Risk Management and Adaptive Protection supply the behavioural dimension — who is doing this, and should protection tighten for them.

A worked scenario

SC-401 is an implementation exam, so it tends to give you a situation rather than a definition. Here is the shape of one, worked through.

Step 1 — establish what actually happened. Copilot does not bypass permissions. If the content appeared, the user had access to it. So this is not a Copilot fault; it is an access fault that Copilot made visible. Use DSPM for AI to confirm the AI interaction and whether sensitive data was involved, and Purview Audit to establish exactly what the user did and when.

Step 2 — find the root cause. Almost always over-permissioning: a SharePoint site or document shared more broadly than intended, or an organisation-wide sharing link. This is where the SharePoint oversharing tooling matters.

Step 3 — apply layered controls. No single setting fixes this:

Step 4 — keep watching. Monitor in DSPM for AI, review Activity explorer for label and DLP activity, and handle alerts in the Purview portal or Defender XDR.

For the full picture of what else changed in the move from SC-400, see SC-400 to SC-401: what changed. To test yourself against current objectives, the SC-401 practice test covers all three domains.

Frequently Asked Questions

What is DSPM for AI?

Data Security Posture Management for AI is a Microsoft Purview capability that discovers and manages how AI services and applications interact with your organisation's data, so you can see AI activity and apply policy to it.

Which SC-401 domain is DSPM for AI in?

It sits in 'Manage risks, alerts, and activities', under the objective group 'Protect data used by AI services'. That domain is weighted 30 to 35 percent.

Was DSPM for AI on the SC-400 exam?

No. The objective did not exist on SC-400, which retired on 31 May 2025. It is one of the clearest additions in SC-401 and a common blind spot for candidates converting from older study material.

Do I need to configure DSPM for AI to pass SC-401?

SC-401 is an associate-level implementation exam, so you should understand prerequisites, roles and permissions, policy configuration and monitoring well enough to apply them, not merely define them. Hands-on time in a trial tenant is strongly recommended.

ExamCert Team

Certified IT professionals tracking the cloud, AI, and security certification landscape. Every exam brief is rebuilt against the official skills-measured document on the date shown above.