If you run a small firm, someone on your team is almost certainly using an AI tool for work already. They draft emails with ChatGPT, summarise documents with Copilot, or tidy up a letter with Gemini. Most of the time nobody decided this. It just started, because the tools are free, quick and genuinely useful. That is not a problem in itself. The problem is that nobody has looked at where the information goes once it is typed in. This post explains where the real risk sits and what a small business can sensibly do about it.
Why banning it rarely works
The first instinct is often to ban AI tools outright. In practice that tends to push the use out of sight rather than stop it. People carry on using the tools on their personal phones and personal accounts, where you have even less say over what happens to the data. A better approach is to decide which tools are allowed, set them up properly, and be clear about what must never go into them. Your team keeps the time it saves, and you know where your data is.
Where the risk actually sits
- Free and personal accounts. On many consumer AI services, what you type can be used to train the provider's models unless the user switches that off in the settings, and it is usually kept for a period even then. Business plans usually exclude your data from training by default, which is a large part of what you are paying for.
- Accounts that leave with the person. If a member of staff uses their own account, the chat history, including anything they pasted in, stays with them when they leave. You have no record of what went in and no way to remove it.
- Copilot and oversharing. Microsoft 365 Copilot shows each person what they already have access to. That sounds safe, but in many firms files have been shared far more widely than anyone realises over the years. Copilot does not create that problem, it just makes it very easy to find. A junior member of staff asking about salaries may be shown the payroll spreadsheet because the folder it sits in was opened up to everyone years ago.
- Confident wrong answers. AI tools can produce text that reads well and is simply wrong: an invented case reference, a misquoted regulation, a figure that does not add up. If it goes to a client unchecked, it goes out under your name.
- Plug-ins and browser extensions. Add-ons that promise to summarise your inbox or write replies for you often need permission to read everything you see. Each one is another company with access to your client data.
What UK GDPR has to say
Pasting personal information about a client into an AI tool is processing personal data, and the usual UK GDPR rules apply. You need a lawful reason to do it, and you should know where in the world the data is processed. The account type matters here. On a business plan the provider normally acts on your behalf and offers a data processing agreement. On a free or personal account there usually is no such agreement, and the provider may use what is typed for its own purposes, which is hard to square with UK GDPR when it is client data. Where a use is likely to be high risk, for example health records or other sensitive personal details, UK GDPR requires a data protection impact assessment before you start, and the Information Commissioner's Office expects one for most uses of AI on personal data. The ICO has published guidance on AI and data protection, and regulators and professional bodies, such as the Solicitors Regulation Authority, the Law Society and the ICAEW, have published material on AI for their members. None of this stops you using AI. It means the choice of tool and account matters, and you should be able to show you thought about it.
Five things to put in place this month
- Find out what is already in use. Ask your team which AI tools they use, on which accounts, and for what kind of work. Make it clear there is no blame. You want an honest answer, not a tidy one.
- Pick the tools you will allow, and use business accounts. Some come with licences you may already have, such as Microsoft 365. A small number of approved tools on business plans, set up by the firm, is far easier to manage than everyone using whatever they found. Check the data settings on each one so your content is not used for training.
- Write a one-page AI policy. Which tools are approved, what must never be typed in, who signs off a new tool, and that a person checks anything before it goes to a client. One page that people read is worth more than ten that nobody opens.
- Tidy up file sharing before switching Copilot on. If you use Microsoft 365, review who can see what in SharePoint, Teams and OneDrive, and remove the Everyone permissions and the open sites and teams that have built up over the years. Do this first, not after someone finds the payroll file.
- Show your team what good looks like. A short session with real examples of what is safe to share and what is not, and why. Most people want to do the right thing. They have just never been told what it is.
What a one-page AI policy should say
- The approved tools, and that work must be done on business accounts, not personal ones.
- What never goes in: client names alongside their details, health or financial records, passwords, anything covered by a confidentiality agreement, and anything you would not want to see on a stranger's screen.
- That a person checks any AI output before it is sent to a client or relied on for a decision.
- Who to ask before trying a new AI tool, plug-in or browser extension.
- What to do if something sensitive has already been pasted in by mistake, and that reporting it will not get anyone into trouble.
Is this a big job?
For most small firms, no. The survey, the settings, the policy and a staff session can usually be done in a few weeks alongside normal work. The Copilot permissions review is the part that grows with the size of your file shares, so it is worth starting early if Copilot is on your plans. After that, the main thing is to look again every so often, because the tools and their settings change quickly.
If you would rather not do it yourself, our Safe AI use service covers the steps above: the survey, the settings, the Microsoft 365 permissions review, a one-page policy written with you, and a short staff session. It is priced per engagement, with a fixed quote agreed before any work begins.
