Your team is already pasting company data into AI.

Somewhere in your business today, someone pasted a client email into an AI chatbot to help draft the reply. Someone else uploaded a spreadsheet to have it summarised. They did it because it saved them twenty minutes, and they were right: it did.

The question isn’t whether your team uses AI. It is which tools, signed in with which accounts, and with what data. For most businesses, nobody knows.

Banning it doesn’t work

The instinctive answer is to block it. It rarely holds. The tools are too useful, and they are on every phone. A ban mostly moves the use to personal accounts, where the business can’t see it, can’t set the terms and can’t get the data back.

The better answer is to give people a tool that is safe to use, and make it the easiest one to reach.

Find what’s already in use

Before you set rules, find out what the rules are for. Four places tell you most of it:

• Expenses and card statements. Individual AI subscriptions show up as small monthly charges.
• Sign-in logs. Your identity platform records every app people have connected to their work account.
• Browser extensions. Many AI tools arrive as an extension that can read every page a person opens.
• Asking. A short, blame-free question — which AI tools do you use for work, and for what? — gets honest answers when people know nobody is in trouble.

The risk is the data, and the access

Two different risks hide under the word “AI”.

The data you send out. Consumer versions of AI tools may keep what you type and, depending on the settings, use it to improve their models. Business versions usually come with contractual commitments about retention and training. The same product can sit on either side of that line depending on how someone signed up.

The access an assistant inherits. Assistants built into your email, documents and chat can see everything the person using them can see. That is the point of them. But it means every folder shared with “everyone” and every link set to “anyone with the link” becomes something a colleague can find by asking a question in plain English. Oversharing that was merely possible becomes easy.

Three rules people will follow

Long AI policies go unread. Three rules, said plainly, get followed:

1. Use the approved tools, signed in with your work account. Nothing else, for work.
2. Never put in customers’ personal data, passwords, or anything under a confidentiality agreement — unless the tool has been approved for exactly that.
3. You own what you send. An email or a report drafted by AI is still yours; check it as if you had written it.

Publish them where people work, alongside the list of approved tools and who to ask when something isn’t on it.

Fix the permissions before you switch it on

If you plan to turn on an assistant inside your email and documents, clean up first. Find the folders shared with the whole company and the links anyone can open, starting with finance, HR and anything holding customer data. Label what is confidential and narrow who can reach it. An assistant switched on over tidy permissions is a productivity tool. The same assistant over years of oversharing is a search engine for things people were never meant to see.

Keep a short register

Keep one list: each AI tool the business uses, who owns it, what data it may see, and who approved it. Review it every quarter, as new tools arrive faster than policies do.

When customers, insurers or auditors start asking how you govern AI — and they are starting to — the register is your answer. If you ever need to show it formally, ISO/IEC 42001 is the standard for managing AI, and the register is where it begins.