Who is liable when AI gets something wrong?
Usually the business that used the AI output. The UK has no AI-specific liability law, so existing rules such as contract and negligence apply.
Why blaming the AI is not a defence
When AI produces a wrong answer, a misleading statement or a flawed piece of work, responsibility does not pass to the software. A business generally remains responsible for the advice, content and decisions it gives clients, whether or not AI was involved.
The UK has no AI-specific liability law. Liability is decided under existing law, such as contract, negligence, consumer protection and data protection. In practice, the same questions apply as for any other mistake: what did you promise, did you take reasonable care, and who suffered a loss?
What a chatbot case shows
In Moffatt v Air Canada (2024), a Canadian tribunal held Air Canada responsible for incorrect information given by the chatbot on its website. It is a Canadian decision, not a UK one, and a UK court would decide a similar case on its own facts. Even so, it illustrates a familiar principle: if your system tells a customer something, the customer will reasonably treat it as coming from you.
Where the risk realistically sits
Most AI errors in an SME are small and caught before anyone outside the business sees them. The realistic risk comes from output that reaches a client or customer without proper checks.
Client-facing advice: A report, recommendation or calculation with an AI-generated error can lead to a claim if the client relies on it and loses money.
Customer service tools: A chatbot that gives wrong prices, terms or policy information can create complaints and commitments you then have to deal with.
Published content: Marketing copy with false claims, or material that copies someone else's work, creates the same exposure as if a person had written it.
Personal data: Errors involving personal information can raise data protection issues on top of any client dispute.
How to reduce your exposure
Keep human review proportionate: Match the level of checking to the stakes. A quick read suits internal notes, but client advice, figures and anything contractual need sign-off from someone qualified to judge them.
Check client contracts: Look at what your terms promise, how liability is limited and whether clients expect to be told when AI is used. Update your templates if they say nothing on the subject.
Check supplier terms: AI providers' terms may limit what you can recover from them, so do not assume you can pass a loss back. Read the terms before you rely on a tool for client work.
Review your insurance: Insurance policies vary in how they treat AI-related claims. Check the wording of your professional indemnity and other relevant cover, and ask your broker directly.
Set clear internal rules: An AI acceptable use policy that makes the sender responsible for checking AI output closes the gap where errors slip through.
Consider EU exposure: If customers in the EU interact with your AI systems, further rules may apply, as covered in whether the EU AI Act applies to UK businesses.
The law is still developing
How courts treat AI-related errors will become clearer as more cases are decided, and much will turn on the facts of each situation. Take advice on your own situation, particularly before relying on AI for regulated advice or customer-facing services.
More questions