15 July 2026

Your team already uses AI (even if you do not know it)

Ask yourself this: if you asked everyone on your team today whether they had used an AI tool for work this week, how many would say yes? Probably more than you imagine. Not because anyone wants to hide anything, but because for many people opening ChatGPT to draft an email or summarise a document is already as normal as opening their inbox.

The problem is not that your team uses AI. The problem is that it does so without anyone ever having talked about how to do it well.

Three real risks (not hypothetical ones)

The first risk is data leaving that should not leave. Someone pastes a client contract, an internal price list or payroll data into a free AI tool to have it summarised or improved. That information has left the company and entered a system you do not control.

The second risk is someone believing an invented answer. Generative AI tools can produce answers that sound confident and are simply wrong: a legal article that does not exist, a made-up figure, a fact that sounds right but is false. If nobody checks the answer before it goes to a client, the mistake belongs to the company, not the tool.

The third risk is less obvious but just as important: everyone uses their own tool, their own way, with no shared standard. One person uses a paid tool with decent privacy terms; another uses the free version of a different tool that does keep and reuse what you type into it. There is no way, as a company, to know what risk is actually being taken.

Why banning it does not work

The instinctive reaction of many small companies is to ban AI at work. In practice that does not remove the risk: it only makes it invisible. People keep using it, but now hidden, without asking questions and without any sense of what is acceptable and what is not. You get exactly the opposite of what you wanted.

What training the team actually means

Training the team does not mean a technical programming course or turning everyone into an AI expert. It means a few clear, shared rules: which data must never be pasted into any AI tool, which tools make sense for the company and which do not, and when an answer has to be checked before it is treated as good and sent to a client.

That is enough for the team to stop using it "on the quiet" and start using it with judgement. And often the result is better than expected: people already knew they had to be careful; they just needed someone to say it out loud and mark out where the lines were.

A real case: a small advisory firm

At a small advisory firm, two staff members had been using free AI tools to draft emails and summarise documents for months, each on her own and without saying so openly. When the manager raised it in a meeting, she realised one of them had, on occasion, pasted client data in to get a quick summary.

They banned nothing. They spent an hour setting three simple rules: never paste client data into any tool, always review the answer before sending it, and use only the tools the company had agreed on. With that, and no investment at all, the biggest risk disappeared and the team kept saving time with AI — now with peace of mind.

Where this connects with the law

Article 4 of the EU AI Act asks organisations using AI to take measures so their people can use it properly. Three written rules and an hour's conversation, recorded with a date, already go a long way towards that — and I wrote up what the article actually requires, without the fear marketing, in the piece on mandatory AI training.

This article is general information, not legal advice for a specific case.