Our HR and finance teams keep hitting an error when they upload sensitivity-labelled documents to a Copilot Studio agent we've published in Microsoft 365 Copilot. Roadmap 571305 says labelled files will soon be supported. What exactly changes, and what should we sort out before it lands?

MRC 571305 vs Learn file-upload docs: Copilot Studio agents on the Microsoft 365 channel will accept MPIP-labelled files with protections enforced. What changes, what stays blocked, and the Steady checks to do first.

Staff can't upload Confidential files to our Copilot Studio agent. Is that about to change, and should we be pleased?

That error is the system doing what it says on the tin, so it’s worth understanding before you celebrate it going away.

Here’s what’s going on today. Learn’s Allow file input from users page for Copilot Studio is blunt: files with encryption, such as sensitivity labels or password protection, aren’t supported. The agent tells the user the file is encrypted and can’t be read, or for some rights-managed files just returns a generic error. Learn’s own suggested workaround is to remove the protection (only if your policies allow it) or use Microsoft Copilot Chat instead, which does support labelled files. In practice the first option is what people do, and that’s the real risk: staff stripping labels to get their work done.

Roadmap 571305 is In development on the MRC, with preview and general availability both targeted for October 2026 (Worldwide; Web). It says users will be able to upload MPIP-labelled files to Copilot Studio agents deployed on the Microsoft 365 channel, that agents will process the content based on the user’s or agent’s account permissions, and that label protections will be enforced end to end. Note the scope: the Microsoft 365 channel, not every channel you might have published to.

What “label protections enforced” means elsewhere in Copilot, per Learn: Copilot only returns content from encrypted items when the user has both VIEW and EXTRACT (copy) usage rights. Labels with user-defined permissions and Double Key Encryption block Copilot and agents from reading the content at all. Where inheritance is supported, new content takes the highest-priority label from its sources. I’d expect the same rules to apply to uploads, but the roadmap doesn’t spell it out and Learn hasn’t been updated yet, so test it rather than assume it.

One phrase deserves a second look: “the user’s/agent’s account permissions”. If an agent runs under its own identity rather than the user’s, whose rights decide what it can read? That’s Schofield’s Third Law in miniature: the easier it is for an agent to open a protected file, the easier it is for the wrong person to get at what’s inside. Ask your maker which authentication the agent uses before you widen access.

Options that actually work:

Fix the label workaround now. Tell staff not to remove labels to get files into agents, and check Purview activity for label downgrades on HR and finance content. Best Steady path, and it costs nothing.
Pilot 571305 with one agent and one cohort. Pick an agent that uses user authentication, upload a Confidential file, a Highly Confidential file and a user-defined-permissions file, and record what happens to each.
Point users at Copilot Chat for labelled files in the meantime. Learn already supports that route, and it keeps your DLP and label rules in play.
Wait and enable for everyone on day one. Quickest, but you’ll find out how agent identity and EXTRACT interact in production rather than in a test.

12 Panes action / comms: Tell Security: 571305 (in development, October 2026) lets Copilot Studio agents on the Microsoft 365 channel accept labelled files; confirm EXTRACT behaviour and agent authentication in a test before rollout. Tell makers: List which of your published agents use their own identity rather than the user’s. Tell HR and Finance: Please don’t remove labels to make uploads work; use Copilot Chat for labelled files until we confirm the change. Leadership: A real blocker is being fixed, and we’ll check it doesn’t open a side door.

Benefits / risks: Fewer staff stripping labels; agents become usable for sensitive teams; protection is meant to travel with the file. On the other side, only the Microsoft 365 channel is mentioned, Learn still says encrypted files aren’t supported, and agent-identity permissions could let content reach people who wouldn’t normally get it.

I’d suggest this week: check Purview for recent label removals on sensitive libraries, list your Microsoft 365 channel agents and how they authenticate, and draft a three-file test plan so you’re ready when the preview appears.

DIGITAL SKILL WORKS LTD / 12 PANES · You ask, we answer