Shadow AI in UK Businesses: What the Latest Warning Means and What to Do Now

Shadow AI is the use of artificial intelligence tools at work without your knowledge or approval, and it is the most significant AI development UK business owners need to deal with this month. A cybersecurity expert warned in a broadcast interview that employees using personal AI accounts for work tasks could leave confidential company information accessible after they resign, exposing employers to data breaches and possible GDPR problems (The European). If your team has laptops, email and a broadband connection, you almost certainly have some shadow AI already.

What happened, in plain English?

On 9 November 2026, Andrius Buinovskis, Vice President of Product Strategy at the cybersecurity firm NordLayer, told Business Matters on The European that companies encouraging AI use are struggling to control the growing number of applications employees bring into the workplace (The European). He described the result as an "AI zoo", with approved and unapproved tools used across finance, sales, design and software development, often without employers knowing what information is being shared.

His central point is the one worth repeating to your management team. When someone uses a personal account rather than a company one, the employer has no visibility and no control over what happens to that information. Nobody is being malicious. People are trying to get their work done faster. But the business carries the risk.

Why does shadow AI matter to UK businesses?

It matters because the information at risk is often the information you are legally responsible for. Customer records, price lists, contracts and staff details are personal and commercial data, and UK GDPR sits squarely on your shoulders as the data controller.

Shadow AI creates four practical risks:

1. Data leaving approved systems. A file uploaded to a personal account may be handled under terms nobody in your business has reviewed. The interviewer noted that provider practices vary and that whether submitted information can be used for model training may depend on the subscription and its settings (The European).
2. Access that outlives the employment. A leaver who used a personal account may still be able to open documents they created or uploaded.
3. Over-permissioned tools. A drafting assistant rarely needs your whole shared drive. Connected tools can receive far more access than the task requires.
4. Convincing but wrong output. AI answers can be invented or steered by hidden instructions in material the tool reads, a risk the NCSC calls prompt injection and highlights as an AI specific concern (NCSC).

What does the NCSC say about AI and cyber security?

The NCSC treats AI as part of your existing security picture rather than a separate universe. Its guidance is aimed at helping organisations understand the risks AI introduces and how to manage them alongside everything else (NCSC). The practical read across for a small or mid sized UK firm is straightforward: know which tools are in use, know what data they touch, and decide deliberately rather than by accident.

That is also why the Australian Cyber Security Centre's AI guidance for small business recommends checking vendor data practices and setting clear rules for what staff may upload (ACSC). It is not UK guidance, but the advice travels well.

How do you find out what AI your team is already using?

Start with a conversation, not a surveillance exercise. Ask each team what tools they use to get work done, and make it safe to answer honestly. You are looking for spreadsheets pasted into chatbots, meeting notes summarised in a personal account, and design or marketing tools nobody approved.

Then check the boring places. Look at which third party apps have access to your email, your shared drive and your accounting system. If you cannot explain why an app has access, that is your first thing to fix. Our data integrations service covers this kind of access review and tidy up, and our code scanning service does the equivalent check on custom software you have had built.

What should you put in place now?

You do not need a 40 page policy. You need five things that actually get followed.

1. One approved route. Give staff a company account on an approved AI tool, funded by the business, so the easy path is also the compliant one. People use personal accounts because nobody gave them a work one.
2. A one page rule. State what may never be uploaded: customer personal data, contracts, payroll, unpublished financials, credentials. Keep it short enough to read.
3. Human review of anything customer facing. Check prices, promises and facts before AI output reaches a customer.
4. A callback step for money and account changes. A familiar writing style or voice should never be enough to authorise a payment change. Confirm through a channel you already trust.
5. A joining and leaving routine. Revoke access on the day someone leaves, and ask what AI accounts they used.

Building AI agents you can actually govern

Approved tools matter, but so does how the AI you build yourself behaves. Our bespoke AI agents service is built so consequential steps, such as sending customer messages or changing records, sit under human review, with each permission decided on purpose rather than granted by default. We are Cyber Essentials certified, which covers a defined baseline of technical controls. It is not a guarantee of security, and no certificate is, but it tells you the fundamentals have been assessed.

What does this mean for your website and hosting?

Most shadow AI risk enters through people, not code. Still, if you run an online store or a client portal, check that any AI feature you have added, or that a plugin added for you, is on a company account and documented. It is the same principle as any other supplier: know who holds your data and why. Our hosting and care plans include ongoing checks, and our team can walk through what is connected to your site.

When is this not a Varsuite job?

If you already have an internal IT team with an AI policy, endpoint monitoring and a procurement process, you probably do not need us for governance. We build and run systems, and we can build AI agents that respect your rules, but we are not your outsourced compliance function. For formal legal advice on UK GDPR, speak to a solicitor or your data protection adviser.

What is the sensible next step?

If you suspect shadow AI in your business, the fastest useful move is a short audit of which AI tools touch your data and what access they hold. We can run that review, close the gaps, and build any AI agent you do need with human review and tight permissions from day one. Get in touch and we will tell you honestly whether you need us.

Frequently asked questions

What is shadow AI?

Shadow AI is the use of AI applications at work without the employer's knowledge or approval, including company staff using personal chatbot accounts for work tasks (The European).

Is using a personal AI account for work a GDPR problem?

It can be. If personal data is uploaded to a tool outside your control, you may struggle to show what happened to it, which sits awkwardly with your duties as a data controller under UK GDPR.

Should we ban AI tools at work?

A ban rarely holds. Guidance for small business points towards clear rules and checking vendor practices rather than prohibition (ACSC).

What is prompt injection?

Prompt injection is where hidden instructions in material an AI system reads cause it to behave unexpectedly. The NCSC highlights it as an AI specific security concern (NCSC).

What should we do first?

Find out which AI tools your team already uses, give them a company approved option, and write a one page rule on what may never be uploaded.