Microsoft 365 Copilot is designed to respect your existing permissions. It only surfaces organisational data a user already has at least view access to, and Microsoft states that prompts, responses, and data accessed through Microsoft Graph aren't used to train its foundation models. The real risk isn't Copilot. It's that most organisations have years of oversharing that nobody noticed until an assistant made everything easy to find.

What Microsoft commits to

According to Microsoft's documentation:

1. Copilot only surfaces data that the individual user can access, using the same access controls as the rest of Microsoft 365.
2. Prompts, responses and graph data aren't used to train the foundation models.
3. Where content is encrypted with Microsoft Purview sensitivity labels, Copilot honours the usage rights granted to the user.
4. Admins can manage retention and search Copilot interactions through Microsoft Purview.

On data residency, Copilot is covered in Microsoft's data residency commitments, including Advanced Data Residency.

Where the risk actually sits

Think of a SharePoint site shared with "everyone" in 2019 for a project that ended long ago. Or a folder of HR spreadsheets with inherited permissions nobody reviewed. Before Copilot, these were technically accessible but practically invisible, because you had to know where to look.

Copilot removes the "knowing where to look" part. Ask it about salaries, restructures or a client's health information, and if a user can technically access a file that answers the question, Copilot can find it.

For Australian organisations handling health, aged care or other sensitive information, that isn't an abstract concern. It's personal information under the Privacy Act.

What to check before you roll it out

1. Find the overshared content. Start with sites and libraries shared broadly across the organisation, and anything with sensitive content and wide access.
2. Apply sensitivity labels to what matters most. Labels with encryption travel with the file and Copilot respects them.
3. Turn on data loss prevention for the categories you care about, such as health or financial information.
4. Start with a small pilot group, watch what gets surfaced, and fix permissions before you expand.

FAQ

1. Does Copilot send our data to OpenAI? Copilot runs on Microsoft's own Azure infrastructure within the Microsoft 365 service boundary, and Microsoft states your data isn't used to train the foundation models.
2. Can we stop Copilot from seeing certain content? Yes. Fixing permissions is the primary control, and sensitivity labels and Microsoft's SharePoint oversharing controls add further layers.
3. How does this relate to AI agents? Agents built on top of Microsoft 365 inherit the same permission model and add new risks when they can act.
4. Want to see what Copilot would surface in your tenant, safely? Ask about our Copilot Business trial: 25 seats, one month, including a permissions review first. Contact us

About the author Timotius Devin is AI Lead at PerData Technology Partners in Melbourne. At PerData he helps Australian organisations design, govern and deploy AI on the Microsoft platform.

Sources

1. Microsoft Learn, Data, Privacy, and Security for Microsoft 365 Copilot: https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-privacy