We’ll send a password for the resource area and regular UK-focused AI updates. Unsubscribe anytime.

Why AI risks matter to a small business

You might think AI risk is something for banks, government departments and large technology companies.

It isn’t.

Imagine you run a small plumbing business.

One employee copies a customer’s email into a free AI chatbot and asks it to write a reply. The email contains the customer’s name, address, details of the plumbing problem and information about their home.

Or perhaps you run a café.

Someone asks AI to create a food description and it invents an ingredient or makes an unsupported claim about allergens.

Or you’re a retailer and ask AI to rewrite product information. It quietly changes one of the specifications.

None of these examples requires a sophisticated AI system.

They’re ordinary uses of AI by ordinary businesses.

And that’s the point.

The biggest AI risks for small businesses often aren’t dramatic futuristic scenarios. They’re everyday mistakes that happen because someone didn’t realise what could go wrong.

The Information Commissioner’s Office says AI can exacerbate existing risks, introduce new ones and make some risks harder to assess and manage. It recommends taking a proportionate, risk-based approach rather than treating every AI use in exactly the same way.

You don’t need to become an AI expert.

You need to know where the danger points are.

1. Data and security risks

For most small businesses, this is the first place to start.

What happens when an employee copies something into an AI tool?

The answer depends on the tool, the account, the settings and the information being entered. That’s why “Is ChatGPT safe?” or “Is this AI tool secure?” is often the wrong question.

The better question is:

“What happens to this particular information when I put it into this particular tool?”

Think about the information your business holds:

• Customer names and contact details
• Addresses
• Employee information
• Invoices
• Supplier contracts
• Pricing
• Business plans
• Passwords and security information
• Confidential customer correspondence
• Photographs
• Financial information
• Information covered by a confidentiality agreement

Not all of this should automatically be entered into an AI system.

The ICO’s guidance stresses security and data minimisation when personal information is processed using AI. In simple terms, don’t collect, share or expose more information than you actually need.

The problem with shadow AI

You might have approved Microsoft 365 Copilot for the business.

But that doesn’t mean your staff aren’t also using other AI tools.

Someone might have a personal ChatGPT account. Someone else might use Gemini. Another employee might find an AI tool online that promises to turn PDFs into summaries.

You may never know.

This is sometimes called “shadow AI”.

The problem isn’t necessarily that the tools are bad.

It’s that you don’t know what information is being put into them or what their terms and settings are.

A simple business rule helps:

“Only use AI tools we’ve approved for business information.”

That doesn’t mean staff can’t experiment with AI.

It means they shouldn’t experiment with customer or company data.

Prompt injection is a newer security risk

If you’re building an AI system that can read emails, websites, documents or other untrusted information, there’s another risk to understand.

Prompt injection is where malicious instructions are hidden in information an AI system reads, potentially causing it to behave in an unintended way.

The UK’s National Cyber Security Centre warns that prompt injection can potentially cause AI systems to reveal confidential information or trigger unintended actions.

You don’t need to know how to defend an AI model against this yourself.

The practical lesson for a small business is simpler:

Don’t give an AI system more access or authority than it needs.

If an AI tool can read your entire customer database and send emails automatically, the consequences of a mistake or attack are much greater than if it can simply draft an email for you to review.

2. Legal and data protection risks

This is where AI can get complicated quickly.

You don’t need to understand every part of UK data protection law.

But you do need to understand one basic principle.

If your business uses AI to process personal information, your existing legal responsibilities don’t disappear.

The ICO’s AI guidance covers lawfulness, fairness, transparency, purpose limitation, data minimisation, accuracy, storage limitation, security and accountability.

“The AI tool is responsible” isn’t a complete answer

Imagine your marketing agency uses AI to write customer emails using information from your customer database.

If something goes wrong, you can’t simply assume:

“The agency used the AI, so it’s their problem.”

Your relationship with the supplier matters.

Your contract matters.

The information being processed matters.

And your own responsibilities as a business matter.

This is why AI should be added to the questions you already ask about suppliers handling customer information.

What about AI training on your data?

This is an area where old AI advice can become misleading.

Not every AI tool treats your information in the same way.

Some business services provide stronger data protections than consumer services. Some let you control whether information is used for model improvement. Others have different terms.

So don’t use a blanket rule such as:

“AI always trains on everything you enter.”

It doesn’t.

But don’t assume the opposite either.

Check the provider’s current terms and privacy documentation for the actual service you’re using.

International data transfers

Your AI provider may process information in different countries.

That doesn’t automatically mean you can’t use the service.

But if you’re processing personal information, international transfers can create additional data protection considerations.

For a small business, the practical approach is to check what the supplier says about where data is processed and what safeguards are used.

If you’re dealing with particularly sensitive information, get appropriate advice rather than guessing.

UK rules are changing

UK data protection law has changed.

The Data (Use and Access) Act 2025 introduced changes in stages, with the ICO confirming on 19 June 2026 that all its data protection provisions were in force.

This is one reason old blog posts about “AI law in the UK” can be unreliable.

Don’t base an important business decision on an article written two years ago.

Check the current ICO guidance.

The ICO’s AI guidance is also being updated following the changes to data protection law.

3. AI can cost you money

AI is supposed to save money.

Sometimes it does.

Sometimes it creates another job for you.

Imagine asking AI to analyse your sales spreadsheet.

It produces a neat summary and tells you that one product category is performing badly.

You change your buying strategy.

Later you discover that the AI misunderstood one column.

The problem wasn’t the cost of the AI subscription.

The problem was the business decision based on an unchecked answer.

AI can create financial risk through:

• Incorrect calculations
• Incorrect financial summaries
• Wrong pricing information
• Invented market statistics
• Incorrect tax information
• Bad forecasts
• Duplicate or incorrect invoices
• Unchecked automated refunds
• Poor purchasing decisions
• Paying for tools nobody actually uses
• Time spent correcting AI-generated work

AI hallucinations

An AI hallucination is simply an answer that sounds convincing but contains incorrect or invented information.

The NCSC specifically warns that generative AI can get things wrong and present incorrect statements as facts.

That matters because good writing can hide a bad fact.

If AI writes:

“Our delivery charge is £8.95.”

It sounds fine.

But perhaps your actual charge is £7.95.

If that appears on your website, you’ve created a problem.

The more expensive the mistake, the more checking you need

A useful rule is:

Low cost of error = light checking.

High cost of error = serious checking.

You don’t need to verify every idea AI gives you.

But you should check important numbers, names, dates, prices, legal claims, technical information and anything that could cost you money.

4. Operational and quality risks

AI can make work faster.

It can also make bad work faster.

That’s an important distinction.

If you use AI to create a first draft, a human can improve it.

If you allow AI output to flow straight into your website, customer emails, stock system or accounts without checking it, a small error can spread quickly.

This is sometimes called a compounding error.

One mistake becomes ten.

Ten becomes a hundred.

AI hallucinations aren’t the only quality problem

AI can also:

• Misunderstand instructions
• Leave out important information
• Change the meaning of text
• Invent details
• Give inconsistent answers
• Produce generic content
• Repeat outdated information
• Make calculations incorrectly
• Misinterpret a customer’s request
• Sound certain when it’s uncertain

This is why “human in the loop” sounds complicated but really isn’t.

It simply means a person checks important AI work before it becomes a real business action.

For example:

AI drafts the customer email.

You read it.

Then you press Send.

That’s human oversight.

Skills can also disappear

There’s another risk that’s easy to overlook.

If people use AI for everything, they can become less confident at doing the underlying job themselves.

A junior member of staff who always asks AI to write customer responses may never learn how to handle difficult customers.

Someone who always asks AI to analyse figures may become less comfortable spotting an obvious error.

The answer isn’t to ban AI.

Teach people the job first, then use AI to help them do it better.

Skills England’s 2026 research highlights the need for workers to develop the skills to use AI effectively, safely and responsibly.

5. Customers, staff and reputation

Your customers don’t necessarily care whether you use AI.

They care about whether you do a good job.

That’s the useful test.

If AI helps you answer an email faster and the answer is accurate and helpful, most customers probably won’t care.

If an AI chatbot gives someone the wrong answer about a refund, an allergy or a service they’ve paid for, they’ll care very quickly.

Don’t hide important AI use

You don’t need to put “This email was written by AI” at the bottom of every message.

But if AI is interacting directly with customers or helping make decisions about them, transparency becomes much more important.

For example, if a customer is dealing with an AI chatbot, don’t make it unnecessarily difficult for them to understand that they’re talking to an automated system.

And give them a sensible way to reach a person when the AI can’t deal with the problem.

Be especially careful with employees

AI can be useful for recruitment, staff training and administrative work.

But decisions about people carry greater risks.

Suppose you ask AI to rank ten CVs.

It produces a list.

You hire the person at the top.

What if the system has picked up patterns that unfairly disadvantage certain candidates?

UK government guidance on responsible AI in recruitment highlights risks including bias, discrimination and digital exclusion.

For a small business, the safest approach is simple:

Use AI to help you assess information.

Don’t let it quietly become the person making the decision.

The same applies to performance reviews, disciplinary matters and other decisions affecting employees.

6. Strategic, supplier and competitive risks

AI risk isn’t just about data and legal problems.

There’s also the risk of making poor business decisions.

Buying AI because everyone else is

You see another business using AI.

You subscribe to three tools.

Then another two.

Six months later you’re paying £150 a month for software nobody really uses.

That’s not an AI strategy.

It’s a collection of subscriptions.

Before buying an AI tool, ask:

• What problem will this solve?
• How much time will it save?
• Who will use it?
• What will it cost?
• What information will it need?
• What happens if we stop using it?
• Can we export our information?
• What happens if the supplier changes its prices?
• What happens if the service goes offline?

Vendor lock-in

If you build a major part of your business around one AI provider, moving away later may be difficult.

For a small business, you don’t need to worry about this every time you use an AI writing tool.

But if you’re building your customer service, sales process or internal systems around one provider, think ahead.

Keep copies of important business information in formats you control.

Don’t build your entire business around a tool you could lose access to tomorrow.

AI can make competitors look more alike

There’s another strategic problem.

If every café uses AI to write its website.

Every plumber uses AI to write the same type of service page.

Every retailer uses AI to produce the same product descriptions.

Everything starts to sound the same.

AI can make your business more efficient.

Your experience, judgement, personality and knowledge are still what make it different.

Use AI to help express those things.

Don’t let it replace them.

7. Your suppliers can create AI risks too

You might have decided not to use AI.

That doesn’t mean AI isn’t being used with your information.

Your accountant might use it.

Your marketing agency might use it.

Your website company might use it.

Your recruitment provider might use it.

Your customer service supplier might use it.

Your software provider might introduce an AI feature without you actively asking for it.

This is why AI should become part of your normal supplier checks.

Ask:

“Do you use AI to process information we provide?”

That’s a perfectly reasonable business question.

Then ask:

“What information does it process and which AI service do you use?”

You don’t need to distrust your suppliers.

You need to know what’s happening to your information.

This matters particularly as more businesses start using AI agents that can take actions rather than simply generate text. The Competition and Markets Authority’s 2026 guidance makes clear that businesses remain responsible for complying with consumer law when they use AI agents.

A simple AI risk policy for a small business

You don’t need a huge document.

You could start with one page.

Your rules might say:

Staff may use approved AI tools for:

• Brainstorming
• Drafting
• Rewriting
• Summarising non-sensitive information
• Creating ideas
• Researching low-risk subjects
• Routine administrative work

Staff must not enter:

• Passwords
• Security codes
• Private encryption keys
• Unnecessary personal information
• Confidential information into unapproved AI tools
• Information covered by a confidentiality agreement unless the approved tool and process allow it

Staff must check:

• Customer-facing information
• Prices
• Numbers
• Dates
• Legal claims
• Tax information
• Safety information
• Product specifications
• Anything that could cause financial or reputational harm if wrong

Staff must not allow AI to make important decisions on its own about:

• Customers
• Employees
• Job applicants
• Refunds
• Complaints
• Contracts
• Financial commitments
• Safety

That is already a much better starting point than having no rules at all.

A 30-minute AI risk check for your business

You can do this without an IT department.

Step 1 — List the AI tools

Write down every AI tool currently being used.

Don’t just list the ones you bought.

Ask staff what they use.

Step 2 — List the information

What information goes into those tools?

Customer information?

Employee information?

Pricing?

Documents?

Emails?

Images?

Step 3 — Identify your biggest risks

Ask:

“What is the worst realistic thing that could happen?”

Don’t imagine a Hollywood cyber attack.

Think about ordinary mistakes.

A customer receives the wrong information.

A confidential document is shared.

A supplier’s AI produces incorrect work.

A member of staff relies on a false answer.

Step 4 — Decide what needs checking

Not everything needs the same level of control.

A social media idea is different from a payroll decision.

A product description is different from a safety instruction.

Step 5 — Write down five rules

Keep them simple enough that staff will actually follow them.

Step 6 — Review the list every few months

AI changes quickly.

New tools appear.

Existing software adds AI features.

Your business changes too.

Don’t write a policy once and forget about it.

The five rules I would start with

If you’re a very small business and all of this feels like too much, start here.

1. Don’t put confidential information into an AI tool unless you’ve checked that the tool is suitable for it.
2. Only use AI tools you’ve approved for business use.
3. Check important AI output before you rely on it.
4. Don’t let AI make important decisions about customers or staff without proper human judgement.
5. Ask suppliers whether they’re using AI with your information.

Those five rules won’t eliminate every risk.

Nothing will.

But they’ll prevent a surprising number of avoidable problems.

Do you actually need AI?

There’s one final risk worth mentioning.

Using AI simply because you feel you should.

You don’t have to use AI.

If you’re a self-employed electrician and AI saves you 20 minutes a week writing emails, that’s useful.

If it takes you longer to check its work than it would have taken to write the email yourself, perhaps don’t use it.

If you’re a small café and a simple spreadsheet does everything you need, you don’t need an AI system analysing your sales every morning.

If you’re a retailer and your product descriptions already work well, you don’t need to replace them just because an AI tool can rewrite them.

The goal isn’t maximum AI use.

It’s better business.

And sometimes the best decision you can make is not to use AI for a particular job.