Who's responsible if my AI tells a customer something wrong?

You are responsible. If the chatbot on your website quotes the wrong price, promises a refund you do not offer, or invents a service you do not provide, that is your problem. Not the software vendor's, and not "the AI's". The feeling that it would land on you is correct, and a tribunal has already said so in writing.

What happened when a company tried to blame the bot

Air Canada's website chatbot told a passenger, Jake Moffatt, that he could book a full-price flight to his grandmother's funeral and claim the bereavement discount afterwards. The airline's real policy said no retroactive claims. When he asked for the refund, Air Canada argued that the chatbot was "a separate legal entity that is responsible for its own actions".

The tribunal rejected that outright. It ruled the chatbot was simply part of Air Canada's website, that the airline was responsible for all the information on it, and that the wrong answer amounted to negligent misrepresentation. Air Canada was ordered to pay CA$650.88 (Moffatt v Air Canada, 2024 BCCRT 149). Small money. Enormous precedent.

That is a Canadian decision, but do not expect a UK court to be kinder. A statement made by a tool you put on your website is a statement made by your business. If a customer reasonably relies on it and loses out, the claim comes to you, and a buried disclaimer in your terms will not impress anyone, least of all where consumers are involved.

Does that mean customer-facing AI is too risky?

No. It means you treat the AI like a new starter who has never met a customer. You would not let them improvise prices on day one, so do not let the bot do it either.

In practice, three rules cover most of the risk:

• Constrain what it can say. The AI answers from an approved source of truth: your real price list, your real policies, your real service area. If the answer is not in there, it says so rather than guessing. Plausible guessing is the failure mode you are designing out.
• Keep the risky moments human. Pricing exceptions, refunds, complaints, anything that smells contractual: the AI's job is to recognise the moment and hand over to a person, not to wing it. We cover those escalation patterns in our guide to AI for customer service email.
• Log everything. Every conversation stored, every handover traceable. If a dispute ever arrives, you want the transcript, not a shrug.

This is the test I give clients: if you would discipline an employee for saying it, your chatbot should not be capable of saying it. That is a design requirement, not a hope.

Where do businesses get caught out?

The off-the-shelf widget that "trains on your whole website" and nothing else. It sounds convenient, but it has no idea which page is current. A 2024 price list still live in a forgotten corner of your site will be quoted back to customers with total confidence. The fix is boring and effective: one maintained source of truth for anything the bot is allowed to state as fact, and a review step whenever prices or policies change.

So yes, the liability is yours, and it always was, the same as it is for your emails, your brochures and your staff. Treat that as the design brief rather than a reason to hide. A properly constrained system is safer than a tired human on a Friday afternoon, because it cannot go off-script. An unconstrained one is Air Canada waiting to happen.