Microsoft Purview Endpoint DLP: GenAI Paste Controls

Summary

Copy and paste is the most common way sensitive data reaches a generative AI prompt, and it leaves no file behind to scan. Microsoft Purview Endpoint DLP closes that gap with the Paste to supported browsers activity. Administrators group risky AI sites, build a rule around sensitive information types, and choose to audit, block with override, or block. This byte walks through that design and the rollout details that keep users productive.

Paste Is the Quiet Exit Path

For this article we will use a company called A.M Finances Group rolled out an AI assistant for its sales team, and within a week analysts were pasting account notes, contract terms, and customer details into public chat sites to get faster summaries. Nothing was uploaded and no file was attached, so classic file-based controls saw nothing. The data moved as plain text through the clipboard and into a web form.

Security teams often discover this only after an audit or a near miss. Banning AI sites rarely works, because people simply move to personal devices, and it throws away real productivity gains. A targeted control that understands content is a better answer than a blanket block.

That is the gap paste controls address. Endpoint DLP watches the paste action itself inside a supported browser, reads the text that is about to land, and decides what to do before the prompt ever receives it. Importantly, the check does not depend on where the text came from. A line copied from a spreadsheet, an email, or a chat window is judged by what it contains, not by the label on the source file.

Group the Destinations First

Paste rules need to know where the text is going. In the Microsoft Purview portal, administrators open the Data Loss Prevention settings, go to Endpoint settings, and expand Browser and domain restrictions to sensitive data. There they create Sensitive service domain groups, which are simple named lists of websites.

A.M Finances Group builds two. One group holds consumer AI sites that should never see customer data. A second holds approved tools that deserve a lighter touch. Splitting destinations this way is what makes graduated enforcement possible later. One detail matters here: the Paste to browser activity works with Sensitive service domain groups and is not supported with the older Service domains setting, so choosing the right list type is the first decision.

Build the Rule Around the Paste

Next comes the policy itself. A.M Finances Group creates a DLP policy scoped to Devices, chooses the option to create or customize advanced rules, and adds a rule with a clear name and description. The condition is content that contains selected sensitive information types, such as national identifiers, payment card numbers, or a custom type for internal account codes.

For the action, the administrator selects Audit or restrict activities on devices and then picks Paste to supported browsers under the service domain and browser activities. The rule is tied to the domain groups from the previous step, so the same detection can behave differently for each destination. Policies can be created in test mode first, which lets the team watch real behavior before anyone is interrupted.

Pick the Right Enforcement Level

Each rule offers three outcomes. Audit records the event and lets the paste through, which is ideal for learning how people actually work. Block with override stops the paste but lets the user continue after giving a business reason, creating a visible speed bump without a hard wall. Block stops the paste outright.

A.M Finances Group starts with audit across every AI site, moves its approved-tools group to block with override once it understands the patterns, and sets its consumer-AI group to block. Alerts and policy tips explain each decision to the user in plain language, so the control teaches instead of surprising. Evidence collection can also capture the matched text, though older antimalware client versions may show scrambled characters until the file is downloaded.

Plan for Browsers and Latency

Coverage depends on the browser. Microsoft Edge is supported on Windows and macOS. Chrome and Firefox are supported on both platforms, but the Microsoft Purview extension is needed on Windows. Safari is supported on macOS. Confirm the extension is deployed before assuming users are covered.

Because the engine classifies text at paste time, users may notice a short delay and occasionally see more than one notification. A.M Finances Group trims the noise by setting the overall action to Audit and listing its blocked sites as exceptions, or the reverse. Keeping the antimalware client, the Edge browser, and Windows updates current also helps. A small pilot group and a week of audit data will show whether thresholds need tuning.

Finally, remember that this control covers pasting into supported browsers on managed devices. It works best alongside sensitivity labels, upload and copy controls, and user education, so each layer covers what the others miss.

Conclusion

Paste controls give A.M Finances Group a practical way to keep generative AI useful while keeping regulated data where it belongs. Group the destinations, tie a sensitive information rule to the paste activity, and match the enforcement level to the risk of each site. Start in audit, learn from the results, and tighten step by step. Done this way, the policy protects data without telling employees to stop using AI.