You ask, we answer · 20 October 2026
Does our Copilot DLP policy cover Cowork, or do we need a second one?
We're piloting Copilot Cowork, which goes off and does multi-step work across mail, files and the web. Our Purview DLP policy was built for Microsoft 365 Copilot Chat. Roadmap 570845 says DLP for Cowork is now rolling out. Will our existing policy apply, and what won't it catch?
Grounded with: Microsoft 365 Roadmap 570845 Microsoft Purview: Data Loss Prevention for Microsoft Cowork (Rolling out, GA October 2026, Worldwide; Web); Microsoft Learn — Learn about using Microsoft Purview Data Loss Prevention to protect interactions with Microsoft 365 Copilot, Copilot Chat, and Cowork; Microsoft Purview service description (DLP for Microsoft Copilot); Microsoft Copilot prompt defense in depth
DIGITAL SKILL WORKS LTD / 12 PANES
Asking this before Cowork goes wider is exactly right. Something that acts on your behalf across mail, files and the web needs the guard rails in place first, not afterwards.
Here’s what’s going on. Roadmap 570845 moved from In development to Rolling out on the MRC this week, with general availability targeted for October 2026 (Worldwide; Web). It says data security admins will be able to use a single policy across Microsoft 365 Copilot and Cowork. Learn’s DLP page has already been renamed to cover Microsoft 365 Copilot, Copilot Chat, and Cowork, and it describes Cowork sitting under the same Microsoft 365 Copilot and Copilot Chat policy location. So the short answer is: you shouldn’t need a second policy, but check the rollout has reached your tenant before you tell anyone it’s covered.
What carries over to Cowork, per Learn:
- Labelled files and emails. A rule on Content contains > Sensitivity labels stops Copilot and Cowork processing the content. The item can still appear in citations, but its content isn’t used.
- Sensitive prompts. A rule on Sensitive information types with Processing prompts means the experience won’t respond, and the prompt isn’t used for internal or web searches. Learn marks this as preview, still rolling out.
- Web search. The same SIT condition with Performing Web Searches blocks external web grounding for that prompt while still answering from internal data. This one matters most for Cowork, because it goes to the web on its own.
What it won’t catch is the bit to put in front of your risk committee. Learn says DLP can’t scan files uploaded directly into prompts; it only checks the text that’s typed. The external email exclusion is described for Copilot and Copilot Chat, not Cowork, so don’t assume it applies. The location also doesn’t support admin units, and policy changes can take up to four hours to show up.
Licensing is uneven. Purview’s service description says DLP for prompts is available to all users of Copilot and Copilot Chat, but blocking Copilot from processing labelled files and emails needs E5-class licensing (Microsoft 365 E5/A5, Purview Suite and similar). If you’re on E3, your Cowork pilot gets prompt protection, not label protection.
Schofield’s First Law, never put data into a program unless you can see exactly how to get it back out, stretches to delegated work too: if you can’t trace what Cowork read and did, you can’t review it later. DLP decides what Cowork can use; it isn’t a record of what Cowork did. Make sure audit and your human-approval rules for its actions are switched on as well.
Options that actually work:
- Extend and test your existing policy. Run a Cowork task that touches a labelled file and a prompt containing a test SIT, in simulation mode first. Best Steady path.
- Add a web-search SIT rule if you don’t have one. It’s the control that best matches how Cowork behaves.
- Tell pilot users about the upload gap. Uploaded files aren’t scanned, so labelled content shouldn’t be dragged in by hand.
- Pause the pilot until GA is confirmed in Learn. Safest on paper, but you lose the chance to tune the policy with real tasks.
12 Panes action / comms: Tell Security: 570845 is rolling out; Cowork uses the Microsoft 365 Copilot and Copilot Chat DLP location; test labels, prompt SITs and web-search SITs in simulation. Tell Licensing: Label-based blocking needs E5-class licensing; confirm who in the pilot has it. Tell pilot users: Don’t upload labelled files into Cowork prompts; DLP doesn’t scan uploads. Leadership: One policy should cover Chat and Cowork; we’re verifying before we widen the pilot.
Benefits / risks: One policy for both experiences, a web-search block that suits an agent that browses, and policy tips you can point at your own guidance page. On the risk side: prompt SIT blocking is still preview, uploaded files aren’t scanned, label protection depends on licence, and the external-email control isn’t documented for Cowork.
I’d suggest this week: confirm the rollout in your Purview portal, run three simulation tests through Cowork, and set the policy tip’s compliance URL to a page that explains your rules in plain English.
DIGITAL SKILL WORKS LTD / 12 PANES · You ask, we answer