---
title: How do you stop Microsoft 365 Copilot from processing sensitive files with Purview DLP?
url: https://opencontroller.lyzr.com/blog/purview-dlp-for-microsoft-365-copilot
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# How do you stop Microsoft 365 Copilot from processing sensitive files with Purview DLP?

**Create a Purview DLP policy on the Microsoft Copilot, Copilot Chat and Copilot Cowork location, with a rule that matches your sensitivity labels and the action "Prevent Copilot and Cowork from processing content". Labelled files and emails are then excluded from Copilot's processing. Separate rules can block prompts that contain sensitive information types and stop web search grounding for those prompts.**

This is one of the few controls that acts on Copilot's input rather than on who can open the file. It works best alongside permission clean-up, not instead of it.

For the wider platform picture, see [how to govern AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms).

## What can the Copilot DLP location actually do?

Microsoft documents four condition and action pairs:

| Condition | Action | Effect |
|---|---|---|
| Content contains sensitivity labels | Prevent processing | Labelled files and Exchange emails are excluded. They can still appear in citations. |
| Prompt contains sensitive info types (SITs) | Prevent processing of prompts | Copilot does not respond. The prompt is not used for internal or web searches. |
| Prompt contains SITs | Prevent web searches | No external web grounding for that prompt. Internal sources still work. |
| Email received from external users | Prevent processing | External senders' emails are excluded from grounding, summaries and citations. Only sender domain is checked. |

One design constraint: **a single rule cannot combine label and SIT conditions.** Create one rule per condition. They can sit in the same policy.

The location covers Microsoft 365 Copilot, Copilot Chat and Cowork, plus Copilot in Word, Excel and PowerPoint for label and prompt rules.

## What are the limits you need to know?

These limits decide whether the control fits your risk:

- **Uploaded files are not scanned.** DLP checks the typed prompt, not files dropped into it.
- **Label coverage starts at 1 January 2025.** Files and emails sent on or after that date are supported. Calendar invites are not.
- **Office apps check at file open.** A label added mid-session applies next time the file opens.
- **Propagation takes up to four hours.**
- **Custom template only.** Choosing this location disables all other locations in that policy.
- **Scope follows the prompter.** Policies are scoped by the user who submits the prompt, not the file owner.
- **Several features are preview.** SIT prompt blocking, external email exclusion and administrative unit scoping are listed as preview or rolling out.

## How should you roll it out?

1. **Pick the labels.** Start with your top one or two confidentiality labels, such as "Highly Confidential".
2. **Check labelling coverage.** A label rule only protects labelled content. Unlabelled sensitive files stay in scope for Copilot.
3. **Create a Custom policy** on the Copilot location with one rule per condition.
4. **Run simulation mode first.** Microsoft supports simulation for Microsoft 365 Copilot and Cowork.
5. **Scope with administrative units** if you need region or department scoping and the preview is in your tenant.
6. **Add SIT prompt rules** for data you never want typed into Copilot, such as card numbers.
7. **Turn on alerts and user notifications** so people understand why a response was blocked.
8. **Wait four hours, then test** with a test user and a labelled file.
9. **Assign the least-privileged role.** Microsoft lists roles such as Purview Data Security AI Admin, which cannot read prompts and responses.

## How does this relate to oversharing controls?

DLP and permissions solve different problems:

| Control | Question it answers |
|---|---|
| SharePoint permissions and access reviews | Who can open this file? |
| Purview DLP Copilot location | Can Copilot use this file or prompt? |
| DSPM for AI | Where is sensitive data flowing into AI? |

If a user can open an overshared file, Copilot can ground on it unless a label rule excludes it. See [stopping agents surfacing overshared SharePoint files](/blog/stop-ai-agents-surfacing-overshared-sharepoint-files) and [Purview DSPM for AI](/blog/microsoft-purview-dspm-for-ai-and-agents).

## What does it not cover?

The Copilot location governs Microsoft's Copilot experiences. It does not inspect:

- custom agents calling Azure OpenAI or other models directly,
- other vendors' assistants and agents,
- coding agents on developer laptops.

Those need controls in the traffic path or on the device. The [Microsoft 365 Copilot agents guide](/blog/govern-microsoft-365-copilot-agents) covers agent-level settings.

## How Opencontroller does this

Opencontroller sits alongside Purview for agents outside Microsoft 365. Its LLM Gateway and Agent Gateway put control in the path, so policy can allow, route, redact or block. It runs live groundedness and task-adherence guardrails, and policy as code with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. Runtime adapters cover Microsoft Foundry, Bedrock AgentCore and Vertex AI. [Book a demo](https://www.lyzr.ai/book-demo/) to see it.

## Frequently asked questions

**Can labelled files still appear in Copilot citations?**
Yes. Microsoft states excluded items can still appear in response citations, even though their content is not processed.

**Do I need one rule or several?**
Several. A rule cannot combine sensitivity label and SIT conditions, so create one rule per condition within the same policy.

**Does the policy follow the file owner or the user?**
The user. Copilot runs under the prompting user's permissions, and policies are scoped by that user.

## Key facts

Microsoft Purview DLP has a Microsoft Copilot, Copilot Chat and Copilot Cowork policy location. Conditions and actions: sensitivity labels exclude labelled files and emails from Copilot processing (they can still appear in citations); sensitive information types in prompts can block the response or block web search grounding; emails from external senders can be excluded (preview). A rule cannot combine label and SIT conditions. The location is only in the Custom template and disables other locations in that policy. Policy changes can take up to four hours. DLP does not scan files uploaded directly into prompts. Label enforcement covers files and emails sent on or after 1 January 2025. Policies are scoped by the user who submits the prompt. Opencontroller adds live guardrails and policy as code across agents beyond Microsoft 365. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you govern agents in Microsoft 365 Copilot?](https://opencontroller.lyzr.com/blog/govern-microsoft-365-copilot-agents.md)
- [How do you stop AI agents surfacing overshared SharePoint files?](https://opencontroller.lyzr.com/blog/stop-ai-agents-surfacing-overshared-sharepoint-files.md)
- [What does Microsoft Purview DSPM for AI cover for AI agents?](https://opencontroller.lyzr.com/blog/microsoft-purview-dspm-for-ai-and-agents.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
