Ask a room of business owners whether their staff use AI at work and most will say “a bit, maybe”. The data says otherwise. Deloitte’s survey of 25,000 UK workers found that 63% use generative AI for work, and almost a third of them do so without their employer knowing. Half have had no training at all.
Someone in sales is using ChatGPT to polish proposals. Someone in finance is pasting spreadsheets into an AI tool to spot trends. Your operations manager has found an assistant that summarises meeting notes. None of them asked permission, because none of them thought they needed to. They’re trying to work faster and do a better job.
That’s a good thing. AI can genuinely save time and money for a small business. But when it’s used without any rules, it creates security and data protection risks that most owners don’t know they’re carrying. Here are the ones I see most often, and what you can do about them.
The risks hiding in everyday AI use
1. Your data may not stay yours. Free and personal AI accounts often keep what you type in, and some use it to train future models. When an employee pastes in a customer contract, a staff list or your pricing model, that information has left your business. You no longer control where it’s stored, who can see it or how long it’s kept.
2. You could breach UK GDPR without realising. If personal data about customers or employees goes into an AI tool, you’re responsible for it. Do you know where that tool processes data, whether it’s outside the UK, or whether you’d be able to delete it if asked? If the answer is “no idea”, that’s a compliance gap, and it’s the kind your customers and the ICO care about.
3. AI tools record more than you think. Many AI services keep detailed activity logs of every prompt, reply and uploaded file, often for troubleshooting. Several recent AI data leaks didn’t involve hackers breaking in at all. The logs were simply left unprotected. In January 2025, researchers at Wiz found a DeepSeek database open to the internet with no password, holding over a million log entries including users’ chat histories in plain text. Researchers have since found millions of exposed chat logs from a Sears customer service chatbot containing customers’ names and addresses. Anything your staff type into an AI tool could end up in a log like this, so treat every prompt as something others might one day see.
4. You can’t protect what you can’t see. “Shadow AI” is the new shadow IT. Staff sign up with their personal email addresses, install browser extensions and often connect apps to company Microsoft 365 or Google accounts. None of it appears on your IT provider’s radar. When someone leaves, their access to those tools, and to any of your data inside them, often leaves with them.
5. AI makes confident mistakes. AI tools can produce answers that look polished but are simply wrong. That matters when the output goes into a quote, a contract, a report to a customer or a piece of code that runs your systems. Someone still needs to check the work.
6. Criminals are using AI too. Phishing emails used to be easy to spot because of poor spelling and odd phrasing. Not any more. Attackers now use AI to write convincing emails, clone voices and impersonate suppliers or directors. Your team’s “that looks a bit off” instinct needs updating.
The next step: AI that acts, not just answers
The tools above mostly answer questions. The next wave, often called Agentic AI, goes further. These AI agents can read your inbox, update your CRM, book meetings, raise invoices or move files on your behalf.
The productivity gains are real, but so is the risk. An agent needs access to your systems to do its job, and it will do what it’s told. That includes instructions hidden inside an email, a web page or a document it’s been asked to read. A cleverly worded message could trick an agent into forwarding sensitive files or approving a payment.
The question to ask isn’t “is this AI clever?” but “what could it do if something went wrong?” Give an agent only the access it genuinely needs, and keep a person in the loop for anything involving money, customer data or external communication.
Five things you can do this month
Banning AI rarely works. Staff just use it on their phones instead, and you lose the benefits as well as visibility. A better approach is to make the safe way the easy way.
- Find out what’s already in use. Ask your team, without blame, which AI tools they use and what for. You’ll learn a lot, and you’ll probably find some good ideas worth backing.
- Choose approved tools. Business versions of the main AI tools usually let you switch off training on your data and manage accounts centrally. They cost a little more, but the control is worth it.
- Write a one-page AI policy. Keep it simple: which tools are approved, what must never be pasted in (personal data, customer confidential information, passwords), and that a person checks the output before it’s used.
- Check your connected apps. Review which apps and extensions have access to your email, files and calendar, and remove anything nobody recognises.
- Update your awareness training. Show staff what AI-written phishing and voice cloning look like. Agree a simple rule: any request to change bank details or make an urgent payment is confirmed by phone on a known number.
AI is an opportunity. Treat it like one.
Your staff aren’t the problem here. They’re showing you where AI can help the business. The job of leadership is to put sensible guardrails in place so you get the speed and savings without handing your data, or your reputation, to someone else.
At RiskVane, we help businesses across Cambridgeshire and Hertfordshire do exactly that. As your Fractional Security Manager, we can review how AI is being used today, set clear and practical rules, and recommend the controls you’ll need as you adopt new tools like Agentic AI.