The NCSC says 71% of UK employees have used AI tools their employer never approved, and that banning them fails. Here's what a smaller business should do instead.
Shadow AI in Small Businesses: Why Blocking It Doesn't Work
Seven in ten UK employees have used an AI tool their employer never approved. Just over half of them do it every week. In a 40-person firm that's around 28 people, and most owners can't name one of the tools involved.
The National Cyber Security Centre put a name to the problem on 7 September 2026 in a piece called The hidden risks of shadow AI. The advice inside it isn't what most business owners expect. The NCSC doesn't tell you to ban the tools. It says the practice won't go away, that policy has fallen behind what employees need, and that the realistic goal is reducing the risk rather than pretending you can remove it.
That's a harder instruction to follow than a ban, and it's the right one.
What is shadow AI?
Shadow AI is the use of AI technology that isn't captured in your organisation's approved systems and processes. Someone pastes a client contract into a free chatbot to get a summary. Someone else runs the month-end figures through an AI tool to find an anomaly faster. Nobody signed anything off, and nobody in the business knows it happened.
It's the same behaviour as shadow IT, the personal Dropbox accounts and unsanctioned project tools we wrote about last year, with one difference that matters. When someone used an unapproved file-sharing tool, your data sat somewhere you didn't control. When someone pastes into a consumer AI assistant, your data may also become training material, and you have no route to get it back.
The NCSC names three consequences: sensitive information and intellectual property leaving the business, losing sight of where your data has gone, and AI agents themselves carrying vulnerabilities an attacker can use.
Why blocking the tools doesn't work
The instinct, when a director hears the 71% figure, is to block everything at the firewall and send a stern email. It feels decisive and it fails within a fortnight.
People turn to these tools because the sanctioned option is slower, worse, or doesn't exist. Block the browser version and they'll use their phone. Block that and they'll email the document to a personal account first. Every control you add without giving people something better pushes the activity further out of sight, which is the opposite of what you wanted. You end up with the same risk and less visibility.
The NCSC's own recommendation is to build a positive security culture, because when people can talk openly about security they're far less likely to route around it. That's a genuine instruction rather than a soft one. It means your team has to believe they can tell you they've been using ChatGPT for six months without it becoming a disciplinary matter. If they can't, you'll never get an accurate picture of your own exposure.
We made this argument about shadow IT in June 2025, before AI made the stakes higher. The cause was speed and convenience then and it's speed and convenience now. Employees usually reach for an unapproved tool to do their job better, and treating that as misconduct gets you nowhere.
The figure your finance director should see
Most coverage of the Microsoft research led on 71%. The number that should hold your attention is smaller and worse.
Almost a quarter of UK staff, 22%, have used unauthorised AI tools for finance-related tasks. That's payroll data, management accounts, supplier details and client billing information, going into consumer services under whatever terms those services happen to publish this month.
Set that next to another finding from the same research. Only 32% of employees are concerned about the privacy of company or customer data they put into consumer AI tools. Two-thirds aren't worried, which means the people handling your most sensitive numbers are the least likely to flag it.
The research behind those figures was carried out by Censuswide for Microsoft in October 2025, across 2,003 UK employees. It's the same study the NCSC cites.
Why this lands harder on a smaller business
The NCSC wrote its guidance for cyber security professionals and large organisations. If you have a security team, shadow AI is a problem they can go and measure on Monday morning.
A 40-person firm has nobody whose job it is to notice. There's no data loss prevention tooling watching what leaves the estate, no monthly review of which cloud services staff have signed into, and no security lead to escalate to. The behaviour is identical to a large enterprise. The chance of anyone spotting it is far lower.
There's a compliance edge to this too. The ICO's position is that in most cases using AI involves processing likely to result in a high risk to people's rights and freedoms, which triggers a legal requirement to carry out a data protection impact assessment. If your staff are running client personal data through tools you've never assessed, that duty is already live and unmet.
Meanwhile, 47% of UK small business owners now use AI tools, more than double the 22% recorded a year earlier, according to Simply Business's survey of 1,842 UK small businesses. Among those who've held back, security and privacy worries are the most common reason at 44%. The businesses most anxious about AI risk are the ones not using it. The risk is sitting with the other half, who are.
What to do about shadow AI this week
Start by finding out what's in use, because you can't manage what you can't see. If you're on Microsoft 365, your admin centre and Defender reporting will show which cloud services staff have signed into with their work accounts. That takes an afternoon and it's usually uncomfortable reading.
Then ask people directly, and ask in a way that gets you an honest answer. A short, no-blame survey asking which AI tools they use and what for will tell you more than any log, because it also tells you why. That "why" is what you need to fix the underlying cause.
Give people a sanctioned tool that's genuinely good enough. This is the step most businesses skip, and skipping it guarantees the policy fails. If the approved option is worse than the free one your team already likes, they'll keep using the free one. Our guide to choosing an AI plan for UK SMBs covers what the paid tiers give you, and the short version is that business tiers keep your prompts out of training data.
Write the policy last, and keep it to a page. It should say yes to something specific rather than listing prohibitions, name the tools people may use and the data they must never paste anywhere, and tell staff who to ask when they're unsure. A policy that only says no gets ignored by the people you most need to reach.
If personal data is involved anywhere in that picture, complete a data protection impact assessment. The ICO expects one and it's the document a regulator will ask for first.
Shadow AI is the visible edge of a gap between what your people need to do their jobs and what you've given them. Close that gap and most of the risk closes with it. If you'd like a clear picture of what your team is currently using and where your exposure sits, book a discovery call and we'll walk through it with you.
