Your chatbot is not a separate legal entity. A tribunal has already said so.

An airline told a tribunal it could not answer for what its own chatbot said, because the chatbot was a separate legal entity responsible for its own actions. The tribunal called that submission remarkable. The reasoning that followed is worth reading in full, because none of it depends on the technology.

1. What the tribunal found

Jake Moffatt’s grandmother died in Ontario on 11 November 2022. That same day Mr Moffatt went to Air Canada’s website to book a flight from Vancouver to Toronto using the airline’s bereavement rates, and while doing so used a support chatbot on the site. The parties accepted that he was not speaking to an employee.

The chatbot’s answer, captured in a screenshot Mr Moffatt put in evidence, told him that if he needed to travel immediately or had already travelled, he could submit his ticket for a reduced bereavement rate within 90 (ninety) days of the ticket being issued, by completing a Ticket Refund Application form.

The words “bereavement fares” in that answer were themselves a hyperlink — to an Air Canada page which said the policy did not apply to requests made after travel had been completed. The chatbot linked to the page that contradicted it.

Mr Moffatt booked on the chatbot’s advice. He flew Vancouver to Toronto for CAD 794.98 and returned for CAD 845.38, a total of CAD 1,630.36. He applied for the bereavement refund on 17 November 2022, comfortably inside the 90 (ninety) day window the chatbot had given him, and then corresponded with the airline through December 2022 and February 2023 trying to get it paid.

On 8 February 2023, an Air Canada representative replied and admitted that the chatbot had provided “misleading words”. The representative pointed to the chatbot’s link to the bereavement travel page and said the airline had noted the issue so that the chatbot could be updated. The matter did not settle, and Mr Moffatt filed his claim.

2. The three arguments Air Canada ran, and why each failed

The claim succeeded in negligent misrepresentation, which the tribunal described as arising where a seller does not exercise reasonable care to ensure its representations are accurate and not misleading. The elements are conventional: a duty of care, an untrue, inaccurate or misleading representation, negligence in making it, reasonable reliance, and resulting loss. The tribunal found a duty of care arose from the commercial relationship between service provider and consumer, and that the standard of care generally requires a company to take reasonable care that its representations are accurate.

Against that framework, the airline advanced three defences.

First, that it could not be liable for information provided by its agents, servants or representatives, including a chatbot. The tribunal noted that Air Canada did not explain why it believed that to be so. What the submission amounted to, in the tribunal’s words, was that the chatbot is a separate legal entity responsible for its own actions — and that a chatbot, while it has an interactive component, is still just part of Air Canada’s website. It should be obvious, the tribunal said, that a company is responsible for all the information on its own website. It makes no difference whether that information comes from a static page or a chatbot.

Second, that the correct information was available elsewhere on the site. The tribunal asked two questions the airline had not answered: why the page titled “Bereavement travel” was inherently more trustworthy than the chatbot, and why customers should have to double-check information found in one part of a website against another part of the same website. There was no reason, it held, why Mr Moffatt should know which section was accurate and which was not. His reliance was reasonable.

Third, that terms in its Domestic Tariff excluded liability. Air Canada made submissions about what the tariff said but did not put the relevant portion of it in evidence. The tribunal observed that Air Canada is a sophisticated litigant which should know that it is not enough to assert that a contract says something without producing the contract. The contractual defence was not proven.

The damages finding is instructive for a further reason. Air Canada was best placed to prove what the bereavement fare actually would have been, and did not. The tribunal drew an adverse inference — assuming the missing evidence would not have helped the airline — and accepted the CAD 380 figure Mr Moffatt had been quoted by telephone. Recalculated with taxes and charges, the loss came to CAD 650.88. A claimed CAD 200 goodwill coupon was disregarded, Mr Moffatt having never accepted it.

3. Why this matters in South Africa

Moffatt is a decision of a Canadian small claims tribunal whose decisions carry no precedential value even in their own jurisdiction. It is not binding on a South African court and it should not be oversold.

What makes it worth reading is what the tribunal did not do. It applied no AI-specific law, drew no distinction between machine-generated and human statements, and engaged in no analysis of the legal status of automated systems. It applied ordinary misrepresentation principles to a new delivery mechanism. Had a printed brochure and a call centre agent contradicted each other, the reasoning would have been identical.

South African law recognises negligent misrepresentation. The Consumer Protection Act prohibits misleading representations to consumers. Neither turns on whether a human or a system produced the statement. A South African business on these facts would be defending the same elements on substantially the same ground — and, on the tariff point, would face the same difficulty if it asserted contractual protection without producing the contract.

There is no dedicated AI statute in South Africa. A Draft National AI Policy was published for public comment on 10 April 2026 and withdrawn on 26 April 2026 after several of its references were found to be fictitious and AI-generated. In the absence of a statute, AI outputs are governed by the law that already exists: POPIA where personal information is involved, the Consumer Protection Act where consumers are, and the ordinary law of delict and contract everywhere else. The absence of an AI statute is not a gap in liability. It is the general law applying, unchanged, to a new tool.

4. The related point on automated decisions

Where a system does more than answer questions, a further provision engages. Section 71 of POPIA gives a data subject the right not to be subject to a decision based solely on automated processing where that decision has legal consequences or similarly significant effects, subject to narrow exceptions. A system that declines an application, sets a price or refuses a service without human involvement falls directly within it.

5. What to do about it

• Treat chatbot output as your own statement. It is a representation by the business, made on the business’s own channel. Govern it the way you would govern a published price list.
• Reconcile your sources — and check your links. The chatbot in Moffatt linked to the page that contradicted it. Where your chatbot, website, terms and staff can each answer the same question, they must answer it the same way.
• Do not expect customers to cross-check you. The tribunal rejected the idea that a consumer must verify one part of a website against another. Accuracy is the business’s job, not the reader’s.
• Constrain the subject matter. Decide what the system may speak about and what it must escalate. Pricing, refunds, eligibility and contractual questions are candidates for escalation.
• Keep a human behind anything that commits you. A quote, a concession, an eligibility confirmation or an undertaking should carry a human signature.
• If you rely on your terms, be able to produce them. Air Canada’s contractual defence failed because the tariff was never put in evidence. Know where your terms are, which version applied, and how you would prove acceptance.
• Keep the evidence you would need. Air Canada lost the quantum argument to an adverse inference. Retain chat transcripts, pricing records and version histories — the party who cannot produce records generally loses the point.
• Disclose that the customer is dealing with a system. Disclosure does not transfer liability, but it manages expectation and supports the reasonableness enquiry.
• Check the deployment against section 71. If the system decides rather than informs, confirm a lawful basis or exception applies and that a route to human review exists.

6. The practical position

The efficiency gains from customer-facing AI are real, and none of this argues against deploying it. The point is narrower. A business cannot put a system on its own channel, invite customers to rely on what it says, and then disown the statement when it turns out to be wrong.

The total award in Moffatt was CAD 812.02. The reason the case is cited around the world is not the amount. It is that a company argued its own chatbot was somebody else, and a tribunal declined to accept it.

Source: Moffatt v Air Canada, 2024 BCCRT 149 (Civil Resolution Tribunal of British Columbia, 14 February 2024). This article is general information for guidance only and is not legal advice. The decision is that of a Canadian tribunal, carries no precedential value and is not binding in South Africa. This article does not create an attorney and client relationship and should not be relied on in place of advice on your specific facts and systems. © DKVG Attorneys. Est. 1921.