A tribunal held Air Canada to its own chatbot's words

The conversation

On 11 November 2022, following his grandmother's death, Jake Moffatt used a support chatbot on Air Canada's website to ask about bereavement fares. The chatbot told him he could book a flight and apply for a reduced fare retroactively, within 90 days, and linked to a separate 'Bereavement travel' page that in fact said the policy did not apply after travel was completed. Relying on the chatbot's answer, Moffatt booked two one-way flights totalling $1,640.36. When Air Canada later refused a partial refund, he took the airline to the Civil Resolution Tribunal of British Columbia (CRT). The tribunal's decision, Moffatt v. Air Canada, 2024 BCCRT 149, was issued on 14 February 2024 by tribunal member Christopher C. Rivers.

What the documents show

The decision records that an Air Canada employee later admitted by email that the chatbot had given 'misleading words.' In the proceeding, Air Canada argued it could not be liable for information from 'an agent, servant, or representative' including a chatbot, in effect treating it as a separate legal actor. The tribunal called this 'a remarkable submission' and rejected it, finding a chatbot 'is still just a part of Air Canada's website.' It found negligent misrepresentation and ordered Air Canada to pay Moffatt $812.02 in damages, interest and fees.

The system boundary

The decision does not describe any escalation the chatbot offered at the time of the conversation; Moffatt only learned the correct policy later, by phone and email with human staff. The CRT itself, as its own description of its mandate states, is 'Canada's first online tribunal,' one of more than twenty administrative bodies resolving disputes 'without a lawyer, and without a courtroom' up to a monetary limit, not a general regulator of chatbots or AI. Its ruling settled this contract dispute on its facts; it did not set rules for other jurisdictions or for claims outside a negligent-misrepresentation theory.

Where it fails

The documented failure is a contradiction the company itself created: a chatbot answer and a linked policy page that said opposite things, with no indication either was flagged as more authoritative than the other. Air Canada's attempted defense also failed for a separate, mechanical reason: it cited its tariff's terms without submitting the tariff itself as evidence. A builder or operator should watch for exactly this kind of divergence between a generated answer and the static page it links to.

• Do a deployed assistant's answers and the pages it links to ever say contradictory things, and how would that surface?
• Who at the company reviews a chatbot's stated policies against the underlying source documents, and how often?
• What evidence would the company need to produce if a customer relied on an incorrect automated answer?

The ruling is narrow: a small-claims tribunal held a company to its own website's words. It does not establish a general rule for other courts, but the reasoning it used is a plain one that any operator publishing an automated assistant should expect to face.