Telling staff not to use AI tools does not stop them using AI tools. It stops them telling you they are.

The instinct to ban AI tools outright is understandable. It feels like the cautious, responsible response to a genuinely new category of risk. Nobody wants to be the business that leaked customer data into a chatbot, or sent a client an AI generated report full of confident errors.

In practice, a ban tends to achieve the opposite of what boards intend. Staff who find AI tools genuinely useful do not stop using them because of a policy. They simply stop being open about it. The business loses visibility into exactly the usage it was trying to control.

The evidence: bans do not stop usage

The research on this is consistent.

A 2026 TrustedTech survey of UK and US employees found that 22% of UK employees said they would keep using AI tools even if their organisation banned them and disciplinary action was possible. The same research found that 76% of UK employees recognise that unapproved AI tools pose security or privacy risks, yet 47% still use them at work.

Interestingly, it is not mainly junior staff. TrustedTech found 62% of UK senior leaders use unapproved AI tools, compared with 31% of employees below decision maker level. The people writing the policies are often among the heaviest users.

Workarounds are easy. A Zapier survey found that around one in four workers whose organisation restricts or blocks AI tools had used a personal device or account to access them anyway. And people often want to keep it quiet: 43% of those using personal AI accounts for work said they wanted to keep that usage private from their employer.

None of this means staff are being malicious. Most are trying to do their jobs faster and better. Zapier’s research found many simply prefer their own tools, or find it too slow or difficult to get an approved one. A ban does not remove that need. It just pushes it out of sight.

The real risk is ungoverned use, not the tools

This matters because the risk with AI tools is not the tools themselves. It is ungoverned use, which typically shows up in three ways.

Sensitive data going into tools with no guarantees. Customer details, staff records, financial information, contract terms and passwords pasted into free consumer tools that may retain the data or use it for training. Zapier found 79% of the workers it surveyed had entered especially sensitive information into AI tools, including HR data, logins and personal data. Much of this happens through personal accounts. LayerX found 67% of generative AI usage runs through non corporate accounts, out of sight of IT entirely.

AI output used without review. AI tools can produce fluent, convincing content that is simply wrong: invented figures, incorrect legal points, made up references. Used without checking, that ends up in client reports, proposals and decisions.

Undisclosed AI influence. Decisions, recommendations or customer communications shaped by AI without anyone knowing. That matters for accountability, for client trust and, in some sectors, for regulatory obligations.

A ban does nothing to reduce any of these risks. It just removes the business’s visibility into where they exist.

Why this is a board issue, not just an IT issue

Ungoverned AI use connects directly to risks boards already care about.

Data protection. Personal data entered into an external AI tool is still personal data. UK GDPR obligations around lawful processing, supplier contracts and security still apply.

Insurance. Cyber insurers increasingly ask what AI tools a business uses and what governance sits around them. “We banned it” is a weak answer if usage is actually widespread.

Client contracts. Many client agreements include confidentiality clauses that pasting their information into a public AI tool could breach.

Governance generally. The government’s latest Cyber Security Breaches Survey found nearly a third of UK businesses are using, adopting or considering AI, but only 24% of those have cyber security measures in place to manage the risks. Most businesses are using AI faster than they are governing it.

What works instead: a proportionate, usable AI policy

The alternative is a proportionate, genuinely usable AI policy. It does not need to be long. For most SMEs, two pages is plenty. What matters is that it is clear, realistic and actually followed.

A good policy covers five things.

1. Approved tools. Name the AI tools staff are allowed to use, and make them easy to access. Ideally these are business versions with proper data protection terms, such as tools that do not train on your data and sit under your organisation’s account and sign on. If the approved tools are good, most people will use them. If they are poor or hard to get, people will drift back to personal accounts.

2. Data rules. Be explicit about what categories of data must never go into an external AI tool. A simple traffic light approach works well:

  • Green: public information, general drafting, research and brainstorming
  • Amber: internal business information, only in approved business tools
  • Red: never in any external tool, such as personal data about customers or staff, passwords and credentials, client confidential material, and financial or commercially sensitive information

3. Review standards. Set a basic rule that AI generated output must be checked by a person before it is used, particularly anything going to a client, informing a decision, or containing facts, figures or legal points. The person who uses it is responsible for it.

4. Disclosure. Be clear about when AI use needs to be disclosed, internally and to clients. For example, when AI has materially shaped a piece of client work, or when a client contract requires it.

5. Review cadence. AI tools and risks are changing quickly. Revisit the policy every six months or so, and whenever a significant new tool or use case appears.

Making it stick

A policy on its own is not enough. Three things make the difference between a document and a working practice.

Training. Short, practical sessions on what is allowed, what is not and why. It makes a real difference: Zapier found 82% of people who felt fully confident about what was safe to share with AI had received formal training, compared with 46% of those who were unsure.

Amnesty and discovery. Before launching a policy, ask staff what tools they are already using and what for, without judgement. You will learn where the real usage is, what needs people are meeting, and which tools are worth approving. Shadow AI is often just an unmet need.

Leadership example. If senior leaders use personal accounts and paste client data into consumer tools, nobody else will take the policy seriously. The rules need to apply at every level.

Governance, not prohibition

In practice, some of the most ungoverned AI use sits in businesses that officially banned it months earlier. The ban created the impression that the risk was handled, while the actual usage carried on, unmanaged and invisible.

Governance, not prohibition, is what actually reduces risk. It gives staff a safe, approved way to get the benefits of AI, and gives the board visibility of where the risks sit and confidence that they are being managed.

If your business has banned AI, or has no policy at all, it is worth finding out what is actually happening before the question comes up on an insurance form or in a client audit.

← All Insights
← Previous

More Insights

Cyber Essentials

The Compliance Question That Trips Up Most SME Boards

GDPR compliance for a SME business is usually a project measured in weeks, not an open ended retainer. Here is what the law actually requires, including the new complaints rule most SMEs have missed.

11 November 2026 · 7 min read
IT Governance

What a Fractional IT Director Actually Does Month to Month

Board level IT oversight without a full time hire sounds good in principle. Here is what the work actually looks like, from month one discovery to the monthly and quarterly rhythm.

6 November 2026 · 8 min read
AI & Technology

How to Stop Staff Putting Sensitive Company Data Through AI Tools

Most businesses have not decided whether their staff should be using AI tools. Their staff have decided for them. Here is what that means for your data.

26 October 2026 · 6 min read