A survey published earlier this month asked over 1,000 senior UK decision-makers whether they thought staff were using AI tools the business hadn’t approved. 88% said yes. Around half said they were worried staff were pasting sensitive information into them.
To be fair, the survey was run by a company that sells AI oversight software, so it was never going to conclude there was nothing to worry about. But it lines up with our findings at many new clients and Microsoft’s own research last year, which found 71% of UK employees had used unapproved AI tools at work. Whatever the exact number, it’s most of your staff.
The industry has settled on calling this “shadow AI”. It’s the same idea as shadow IT - people using whatever gets the job done, whether or not anyone signed it off - except these AI tools are now very good at swallowing large amounts of your data if left unsupervised.
Your staff are almost certainly using AI already, and banning it won’t stop them. Give them an approved tool, a one-page set of rules, and stop third-party apps connecting to your company data without someone checking first.
And yes, this applies to directors too - who usually have the most sensitive inbox in the building.
What shadow AI actually looks like
It’s rarely anything dramatic. Usually it’s someone trying to be efficient:
- Pasting a long client email chain into a free chatbot to draft a reply
- Uploading a spreadsheet of customer data and asking for a summary
- An AI note-taker that joins every Teams call and emails the transcript to an outside service
- A browser extension that “helps you write” by reading every page you open
- An AI feature quietly switched on inside an app you already pay for
There’s also a newer category that worries us more: AI assistants and agents that staff connect directly to their mailbox, files or calendar with a “Sign in with Microsoft” or “Sign in with Google” button. That’s not someone pasting in one email - it’s ongoing access to everything in that account, often with nobody else aware it exists.
Then there are the leftovers. A tool someone connected for a project last year, which nobody uses anymore, but which still has permission to read their files. The person who set it up may even have left the business.
Why it matters
Your data leaves your control. Once it’s typed into a free consumer AI tool, it’s on someone else’s servers under someone else’s terms. Depending on the service and its settings, it may be kept and used to train future models.
You’re still responsible for it. If personal data about a client or colleague goes somewhere it shouldn’t, that can be a data breach under UK GDPR, and a serious one needs reporting to the ICO within 72 hours. “A member of staff put it into an AI tool” is a valid explanation, but it doesn’t work so well as a defence.
Your client contracts may say no. Plenty of NDAs and client agreements restrict where information can be processed. A free chatbot almost certainly isn’t on the approved list.
The output isn’t always right. AI tools are very much like your other half in an argument: confident even when they’re wrong. A badly summarised contract or a made-up figure in a client report is a reputational problem, not a technology one.
Our take
Blocking AI tools on the office network achieves very little. People just use their phones, and now you can’t see it at all.We’d much rather give staff something approved that works well, and be clear about what never goes into it. Most people aren’t trying to cause trouble - they’re trying to clear their inbox before finishing for the day.
What to do about it
Ask. A short, no-blame question to staff about which AI tools they use and what for will tell you more than any software. You’ll probably also learn which tasks are worth helping them with properly.
Pick an approved tool. If you’re on Microsoft 365 or Google Workspace, there’s already an AI assistant included - Copilot Chat or Gemini - that works with staff’s work accounts and doesn’t use your data to train its models. That’s a sensible free and easy starting point for most businesses. The paid business plans from the main AI providers offer similar terms if you want more powerful options.
Write the rules down - on one page at most. Which tool to use, what never goes into any AI tool (client personal data, financials, passwords, anything under NDA), and a reminder to check the output before it goes anywhere. If it’s longer than a page, nobody will read it.
Stop apps helping themselves. By default, many Microsoft 365 and Google Workspace setups let users grant third-party apps access to their mailbox and files without anyone checking. Whoever manages your email and files can turn that off and switch on an approval process, so new requests come to someone who can say yes or no. It’s one of the most useful ten-minute changes most businesses haven’t made.
Review what’s already connected. Whoever manages your email and files can also give you a list of the apps already connected to your company accounts. Most businesses are surprised by what’s in there. Anything nobody can explain, or that has no owner, should be disabled.
And it’s not just email. Your accounts software, CRM, chat and file-sharing tools probably have their own list of connected apps, and increasingly their own AI features switched on by default. They’re worth the same look.
Do it again in three to six months. New tools are appearing faster than ever, and each release is more capable than the last. Staff come and go. This isn’t a one-off job.
Private clients and family offices
If you run a family office, all of the above applies with the volume turned up - plus a few things most businesses never have to think about.
The family, not just the staff. Principals and family members often use consumer AI tools on personal phones and accounts that sit entirely outside the office’s own systems, so none of the controls above reach them. For private clients without an office behind them, that’s the whole picture - and it’s worth a conversation at home about what goes into those tools.
The data is different. For most businesses a leak is a data protection problem. For a family, travel plans, locations, property details and children’s schools are a personal security issue.
Assistants under pressure. The people managing diaries, travel and correspondence are under real pressure to do it quickly, which makes them the most tempted to hand some of it to an AI assistant. Connect one of those to the principal’s mailbox or calendar and it has access to the most sensitive account in the organisation.
The wider circle. Lawyers, accountants, and other advisers all handle the family’s information. It’s worth asking them what their own AI policies are - and noticing how they answer. Any contracts including those with household staff should document appropriate restrictions.
Hayes IT can audit which apps have access to your business data, set up the approval process, and give you a one-page AI policy template to start from - one that people will actually follow. If you’d like a hand, get in touch.