Last updated: 19 June 2026
Artificial intelligence is everywhere now — in the email client, in customer service, in the tools we use every day. Most people focus on what the tools can do, and that is understandable: that is where the immediate value lies. Far fewer ask the question that actually matters for a business over time: what happens to the data you feed in?
This is not about being sceptical of AI, and it is not an argument for staying away. It is about adopting the tools with your eyes open, so that you know what you have actually said yes to.
Your data is part of the equation
Most AI tools run with an external provider. When you type something in — a customer enquiry, an internal note, a draft agreement — it is sent to their servers, processed there, and the answer comes back. For much of what we do that is entirely unproblematic. But it means that part of what you send leaves your control, and what happens next is governed by the provider's terms, not by you.
It is worth remembering an old principle that still holds: if the service is free, it is often because something other than money is the payment. It could be your data, your attention, or the right to use what you feed in to improve the service. That does not make free tools bad — but it makes it worth reading what you are actually giving away.
This is not necessarily a problem. It is a factor you should be aware of, so that you can choose deliberately what you send where.
A concrete example
Imagine you use the same AI tool for two tasks over the course of a day.
The first: you ask it to write a blog post about something you have already published publicly. There is little that can go wrong here — the content is meant for the world anyway.
The second: you paste in a confidential client contract and ask for a summary. Now you have sent a document containing trade secrets and perhaps personal data out of the building, to a server you do not control, governed by terms you probably have not read closely.
Same tool, same user, same day — but completely different risk. The difference is not in the technology. It is in what kind of data went in.
The questions worth asking
Before an AI tool is adopted in the business, a few simple questions settle most of it:
- What kind of data are we sending in? Public and harmless, or sensitive and business-critical? This is the most important question, and it governs all the others.
- Where is the data stored, and for how long? Within the EU/EEA or elsewhere? Is it deleted, or does it stay?
- Is what we send in used to train future models? Some providers do this by default unless you actively turn it off.
- Who else can get access to it? Subcontractors, employees at the provider, authorities in the country where the servers sit?
- What happens if the service changes? Terms, price and access can change — how vulnerable does your working day become if that happens?
The answers often decide not whether you should use the tool, but how and for what. There is a difference between letting AI help you word a public post and pasting in something you would otherwise lock in a cabinet.
Not everything is equally sensitive
The most practical move is to let the type of data govern the use. Much of what a business does with AI is harmless, and there the ordinary cloud-based tools are excellent — fast, cheap and powerful. There is no reason to make simple things complicated.
For what is sensitive, it is worth thinking twice: whether it should be sent out at all, whether it can be anonymised first, or whether it should be handled a different way. Some choose to run models locally for precisely that kind of data, so that nothing leaves the building. That is one of several options, and the right choice depends on what kind of data it actually is.
Simple ground rules that actually work
You do not need a thick policy to be on safe ground. For most businesses a few concrete rules that everyone understands are enough:
- Decide which categories of data should never be pasted into an external AI tool — typically personal data about customers, contracts and anything marked confidential.
- Be clear about which tools are approved for what, so that each employee does not quietly find their own solution.
- Turn off sharing for model training where the provider offers it, and document that it has been done.
- Talk about it openly internally. Most leaks do not happen out of bad intent, but because someone did not know that something was sensitive.
The good thing about rules like these is that they cost nothing to introduce and make the simple choice the default one.
Dependency is worth a thought
A final point that is often overlooked: if you build a fixed work process on top of an external service, you also become dependent on that service continuing on the terms you planned for. Terms, prices and access can change for reasons that have nothing to do with you.
That is not an argument against using cloud services — the vast majority of businesses do, and with good reason. It is a reason to know what you are basing yourself on, and not to put all your eggs in a basket you do not control. The more critical a process is to operations, the more it is worth having thought through what the plan is if the tool one day looks different.
Awareness is the whole point
You do not need a definitive answer to get started with AI. You need to have asked the questions above, so that your choices are deliberate rather than accidental. Most mistakes in this area do not happen because someone made a bad choice, but because nobody made a choice at all — the data just seeped out, one paste at a time, without anyone having decided that it was fine.
If you are considering adopting AI in your business, it is worth going through these questions before you settle on a tool. It costs little, and it is far easier to think through in advance than to clean up afterwards.