Can Microsoft Purview Actually Stop Staff Pasting Client Data Into ChatGPT?

Detailed Explanation

Purview is not one feature — it's a stack of separate controls, and the honest answer to "will it stop staff pasting client data into ChatGPT" depends on which layer of that stack you've actually turned on, which browser the prompt goes through, and whether the device is one your business manages. Most vendor conversations compress all of that into "yes, Purview handles shadow AI," which is true in the way "yes, a car has brakes" is true of a car with no brake pads fitted.

The layer that does real-time content inspection — reading the text a user is about to submit to an AI site and blocking it before it leaves the browser if it matches a sensitive information type or label — is called Browser Data Security, and as of Microsoft's current deployment guidance it works inside Microsoft Edge only. It covers consumer ChatGPT, the consumer version of Copilot, Google Gemini, and DeepSeek specifically. If your business has standardised on Edge for business use, this is a genuinely strong control: it inspects the prompt itself, not just the destination, and blocks the submission before any data leaves.

Step outside Edge and the picture changes. Purview's Network Data Security extends visibility to "non-Microsoft browsers, apps, APIs, add-ins, and more," but Microsoft's own documentation describes its role there as detecting data leakage, not preventing it — a Chrome or Safari user pasting the same sensitive text into ChatGPT generates a log entry, not a block. For a business relying on this layer alone outside Edge, the control is an audit trail you review after the fact, not a gate that stops the paste from happening.

What Actually Blocks Outside Edge

Endpoint DLP is the piece that can close some of that gap, and it works differently to Browser Data Security — instead of inspecting content inside a specific browser, it watches actions at the device level: paste to browser, copy to clipboard, upload to a cloud service. Point an Endpoint DLP policy at AI application websites and it can block those actions on a managed device regardless of which browser is open, run in simulation mode first to check for false positives, then switch to enforce.

The catch is the word "managed." Endpoint DLP requires the device to be under your organisation's management — in practice, enrolled in Intune or an equivalent — and licensed for it. A staff member's personal laptop, a contractor's own machine, or a managed device where nobody has actually built the AI-website policy yet all sit outside this protection, even though the marketing conversation about "Purview stops shadow AI" implies otherwise.

Sensitivity labels add a third, different kind of protection: content encrypted under a Purview sensitivity label can't be decrypted by an AI app it's pasted or uploaded into, which is a strong control for labelled files but does nothing for a staff member who simply retypes the confidential figure into a chat box from memory — no technology stops that path.

Licensing You'll Actually Need

The full combination — DSPM for AI for visibility, Endpoint DLP for device-level blocking, and Browser Data Security for Edge-specific prompt inspection — generally sits behind Microsoft 365 E5, Microsoft 365 E5 Compliance, or the standalone Purview add-ons layered onto a lower base licence (Business Premium tenants can reach much of this through the Purview Suite add-on). A narrower version of DSPM for AI is also bundled with E3 plus a Copilot licence, but that combination is weighted toward reporting on AI usage rather than blocking it in real time.

Licensing detail in this space changes often enough, and Microsoft's own pages don't always agree with each other on edge cases, that it's worth confirming the exact requirement against your tenant's current SKU — through your Microsoft partner or the Microsoft 365 admin centre — before telling a client or your own board that a specific control is covered.

Things to Consider

• Edge adoption is now a security decision, not just a browser preference. If your business genuinely wants real-time prompt blocking rather than after-the-fact detection, standardising on Edge for business use is part of how you get it — not a side detail.
• "Managed device" has to mean something concrete. Endpoint DLP's protection only extends as far as your device management actually reaches. If BYOD is common and those devices aren't enrolled, the blocking layer doesn't apply to them at all.
• Detection still has value even where blocking doesn't apply. Network Data Security's non-Edge visibility won't stop the paste, but it tells you it happened — which is enough to have the conversation with the employee and tighten the gap, and it feeds the kind of record covered in how do you stop employees from using unauthorized AI tools.
• Run new blocking policies in simulation first. Endpoint DLP and Browser Data Security policies both support a simulation or audit-only mode before you switch on enforcement — skipping that step is how a well-intentioned rollout turns into a flood of help-desk tickets from legitimate work blocked by an over-broad rule.

Common Mistakes

• Assuming "we have Purview" means blocking is switched on everywhere. Purview is licensed and deployed in layers; having access to the platform is not the same as having Endpoint DLP policies built, Browser Data Security configured, and Edge actually the default browser on every managed machine.
• Treating detection in non-Edge browsers as prevention. A Network Data Security alert that fires after a Chrome user has already pasted sensitive data into ChatGPT is useful for follow-up, but it's not the same control as a block — don't represent it as one in a client-facing security answer or a board report.
• Forgetting unmanaged devices exist. A rollout plan that only accounts for company-issued laptops misses contractors, BYOD staff, and anyone working from a personal device — for those, none of this stack applies until the device is brought under management.
• Buying the licence before scoping which control you actually need. E5 is a significant cost jump from a lower Microsoft 365 tier; confirm which specific capability (visibility, device-level blocking, or Edge prompt inspection) solves the actual risk before committing to it, since a narrower Purview add-on may cover the genuine gap for less.