What is shadow AI?

Shadow AI is any AI tool being used for work without the business having approved it, set it up, or - in most cases - even heard of it. The marketing exec running client copy through a personal ChatGPT account. The ops manager pasting a supplier contract into a free summariser. The finance assistant using an AI plugin to build spreadsheet formulas. The new starter recording meetings with a free transcription tool because typing notes is slow.

None of this is malicious. Most of it is initiative. The name borrows from "shadow IT" - the unapproved software that's been creeping into businesses since someone first expensed a Dropbox account - but the AI version moves faster, because the tools are free, they're everywhere, and they're useful from the first minute.

How widespread is shadow AI?

More widespread than you'd guess. Microsoft's UK research in late 2025 found that 71% of UK employees had used unapproved consumer AI tools at work, and 51% were doing so every week. Its global Work Trend Index put the share of AI users bringing their own AI tools to work at 78% - rising to 80% in small and medium-sized businesses.

Scale that to a 40-person business and the picture gets concrete: perhaps 28 people using AI tools you've never heard of, 20 of them every week. That's not one rogue employee. That's a workforce quietly re-tooling itself. Whatever the exact number in your business, the direction is clear: if you haven't deliberately addressed this, you're almost certainly not the exception.

The same UK research offers a clue as to why. 41% of people said they used consumer AI tools because they already knew them from home. 28% said their company simply didn't provide an approved alternative. In other words: your people aren't going around your systems. In most businesses, there's no system to go around.

Why is shadow AI a real business risk?

The instinct is to file this under IT. It isn't. The risk is commercial.

Start with data. When someone pastes a client contract, a price list, or a set of customer records into a free public AI tool, that information has left the building. You don't know where it's stored, what it trains, or who might see it. For a business with client confidentiality clauses, contractual data obligations, or anything resembling a trade secret, that isn't hypothetical. And if personal data is going in, you may have a UK GDPR question you didn't know you'd been asked. It's an exposure you can't measure, because you can't see it.

Then there's accuracy. AI output is confident whether it's right or wrong, and shadow use means nobody has agreed who checks it. If a proposal, a quote or a compliance document goes out with an AI-generated error in it, the client doesn't care which tool made the mistake. Your name is on it.

And there's continuity. When useful work quietly depends on one person's personal AI account - their prompts, their history, their way of working - that capability walks out the door when they do. You can't hand over what you never knew existed.

None of this is an argument for panic. It's an argument for visibility. A risk you can see is a risk you can manage. This one is invisible by definition.

Why banning AI tools doesn't work

The reflex is to ban it. It doesn't work, for the same reason blocking personal email never did - people route around it to get their jobs done. A ban doesn't remove the risk; it drives it further underground, where you have even less visibility than you started with.

There's a quieter cost too. The businesses getting real value from AI right now are the ones learning, week by week, how to use it well. A blanket ban doesn't just push usage out of sight - it hands that learning to your competitors.

Remember why people are doing this in the first place: the tools are familiar, and in many businesses there's no approved alternative. A ban answers neither of those. The honest read is that your people reached for these tools because they help. The job isn't to stop them. It's to give them a safe way to do what they're already doing.

What should you do about shadow AI?

Three steps, none of them dramatic.

1. Find out what's actually being used

Ask your team - without blame. Make it explicit that nobody's in trouble, then ask what tools people use, for what tasks, and what goes into them. You'll learn more in an honest afternoon than in any audit. Owners who do this tend to be surprised twice: first by how much is in use, then by how useful most of it turns out to be.

2. Give people a sanctioned route

Pick an approved tool and set it up properly - a business account, with data protection settings that mean what goes in stays yours. This costs less than you'd expect, and it removes the main reason shadow AI exists: more than a quarter of people in Microsoft's survey used consumer tools simply because there was no approved alternative. Give them one, and much of the shadow clears on its own.

3. Write the simple rules down

One page. What's fine, what's never okay - customer personal data, financials, anything under NDA - and who to ask when it's unclear. Keep it to a page precisely because a policy nobody reads is the same as no policy. Most people follow clear guidance when it exists. The problem is almost always that it doesn't.

The same problem, wearing new clothes

Shadow AI is a new face on an old problem: tools entering the business faster than anyone's tracking them, with no single person owning the whole picture. It's the software bill creeping up, the duplicate subscriptions, the integration nobody understands - the same clarity problem, wearing this year's clothes.

That's also why the fix isn't really an AI fix. Get visibility over your technology estate - every system, tool and subscription the business runs on - and shadow AI stops being a blind spot and becomes just another thing you manage on purpose.