Copilot will find your payroll spreadsheet

There's a conversation we've had a dozen times this year. It starts with a business owner asking whether they should buy Copilot licences, and it ends with us running a permissions report and everyone going quiet for a bit.

The pitch for Copilot is that it can answer questions using your own company's documents. That's also the problem. It answers questions using your own company's documents — all of the ones the person asking is technically allowed to open. Not the ones you'd expect them to open. The ones the permissions actually allow.

For most businesses those two sets are very different, and nobody has ever had a reason to find out by how much. Until now it didn't matter much, because finding the payroll spreadsheet required knowing it existed, knowing roughly where it lived, and being bothered. Copilot removes all three of those obstacles. You just have to ask a question that a payroll spreadsheet happens to answer.

Why the permissions are a mess

This isn't a criticism of whoever set your tenant up. It's how SharePoint works by default, and a decade of small reasonable decisions.

Microsoft's own documentation is blunt about the starting position: by default, SharePoint sets sharing to the most permissive option. From there, the mess accumulates in predictable ways:

• "Everyone Except External Users". A permissions group that sounds cautious and means "the whole company". It gets used once to unblock someone quickly and then sits there for six years.
• Broken inheritance. Someone needed access to one folder inside a locked-down site, so the folder's permissions were detached from the parent. That folder is now invisible to every permissions review anyone does by eye.
• Sharing links that never expired. A link sent to a subcontractor in 2021 that still works today, because nobody set an expiry and nobody could have told you it existed.
• Ownerless sites. The person who created the site left the business. The site didn't.
• Oversized audiences. Sites shared with hundreds of people because it was easier than working out which forty actually needed it.

None of that was reckless at the time. It's just that the consequences were theoretical, and now there's a very fast, very literal assistant sitting on top of it.

What we actually do about it

The good news is this is measurable rather than mystical. Microsoft ships the tooling to find it.

The Content Management Assessment in SharePoint Advanced Management surfaces exactly the list above — oversized audiences, EEEU usage, broken inheritance, inappropriate sharing, and sites that are inactive or have no owner. Microsoft recommends re-running it every 30 days, which tells you something about how quickly it drifts back.

Alongside it there are data access governance reports: a site permissions baseline, and a report on sharing link activity. Between them you get an actual inventory rather than a feeling.

Then there's the control that buys you time. Restricted Content Discovery excludes sensitive sites from Copilot's discovery without changing their permissions at all. It isn't a fix — the underlying access is still wrong — but it means you can deploy Copilot on Monday without waiting for a six-week clean-up to finish. Use it as scaffolding, not as the building.

Underneath all of that sits the slower work: sensitivity labels so documents carry their own classification, Purview data loss prevention so labelled content stays out of Copilot's grounding, expiry on external sharing, and an owner attached to every site. One thing worth knowing if you're an older Microsoft shop — legacy IRM-protected documents aren't used in Copilot grounding at all, so if you protected things that way years ago they need migrating to sensitivity labels before they'll behave the way you expect.

The licensing bit nobody mentions

We mention this early in every conversation, because it's the thing that separates an honest answer from a sales pitch. There is a real version of this problem that can be worked on without the advanced tooling — manual review, tightening the default sharing settings, expiry policies, cleaning up the obvious offenders — and there's a version that needs a licence change. Which one you're in is worth knowing before anyone quotes you.

Do the clean-up anyway

Here's the part that makes this easy to decide. Every bit of the work above is worth doing whether or not you ever buy a single Copilot licence.

Sharing links that never expire are a problem on their own. So are ownerless sites, so is company-wide access to the finance folder, and so is a subcontractor from three projects ago still holding a live link to your drawings. Copilot didn't create any of that. It just made it fast to find.

So if someone's trying to sell you a "Copilot readiness programme", the honest framing is: it's a permissions clean-up, it was overdue anyway, and the AI is the reason you finally got budget for it.

One more thing, on staff using AI on their own

Separate problem, same week. Whether or not you've deployed anything, some of your people are already pasting work into whichever chatbot is open in another tab. The OAIC's guidance on commercially available AI products recommends organisations don't enter personal information — particularly sensitive information — into publicly available generative AI tools, and notes that where sensitive information is involved consent is generally required, and that consent can't be implied just because someone was notified.

Banning it outright tends to push it underground. A short written policy, a sanctioned tool that's actually good enough to use, and a conversation about what should never be pasted anywhere gets you much further.

If you want to know what your own tenant looks like, a permissions review is part of the free IT review. It usually takes a couple of hours and the findings tend to be the most interesting thing anyone shows you that month.

More on Microsoft 365 and AI readiness  ·  Book a free IT review