How do we stop people using AI tools we have not approved?

With three things together: an approved AI tool that does the job, a short policy that says what is allowed, and the security tools in Microsoft 365 to find and block the rest. Blocking on its own sends people to their phones, and a policy on its own gets ignored. In Microsoft 365, Defender for Cloud Apps shows which AI apps are in use, marking an app as unsanctioned blocks it on devices onboarded to Defender for Endpoint, and Microsoft Purview can stop sensitive data going into the AI sites you do allow.

Start with a tool people are allowed to use

People reach for AI because it saves them time. Take that away with nothing in its place and they find another way. Copilot Chat on a work account is included at no extra charge with eligible Microsoft 365 business subscriptions, and its chats get the same data protection as your email. That makes it a sensible default before you block anything else. Are my AI chats safe? explains what that protection covers.

Write down what is allowed

A page is enough. It should name the approved tools and the account to use with each, list what never goes into any AI tool (client data, personal data, passwords, anything under contract), and say who approves a new tool and how to ask. Without it, blocking an app looks arbitrary, and nobody knows how to get the tool they actually need.

Find out what is already in use

You cannot block what you cannot see. Microsoft Defender for Cloud Apps keeps a catalogue of cloud apps with a risk score for each, and it has a Generative AI category. Its cloud discovery shows which of those apps your people are actually using. Microsoft's own guidance suggests a policy that alerts you whenever a new generative AI app turns up (Microsoft Learn).

Look at your app consents too. In Microsoft Entra, the enterprise apps list shows every app a user has allowed to read data with their work account. AI note-takers and email assistants often arrive this way, with no download at all.

Block what you have not approved

When you mark an app as unsanctioned in Defender for Cloud Apps, Microsoft says it is "automatically blocked across devices that are onboarded to Defender for Endpoint". You can choose to warn and educate users instead of blocking outright, which suits a business still rolling out its policy (Microsoft Learn).

For apps that connect to your data, tighten the Entra user consent setting so users can only consent to apps from verified publishers, and switch on the admin consent workflow so they can ask for anything else (Microsoft Learn).

Protect the data, not only the door

Some AI tools you will allow, and some people will still try to paste in things they should not. Microsoft Purview’s Data Security Posture Management for AI shows how people use AI apps and what sensitive information goes into them, and comes with policies you can switch on. Data loss prevention in Microsoft Edge can block sensitive information from being sent to AI sites. Microsoft’s guidance pairs it with blocking other browsers on managed devices, so the protected browser is the only way in.

What this does not cover

All of this works on devices and accounts you manage. Nothing in Microsoft 365 stops someone typing a client’s problem into an app on their personal phone. That is why the approved tool and the written policy come first: the technical controls catch mistakes and shortcuts, and the policy and the training deal with everything else. Which of these tools you already own depends on your Microsoft 365 plan.

The short version

If you want to know what AI is already in use before you decide what to block, our AI and agent readiness assessment inventories every AI system and agent in your estate and gives you a prioritised plan to govern them.