Air Canada Said Its Chatbot Was a Separate Entity. A Tribunal Charged It $812.

The cheapest legal lesson in AI costs $812.02 Canadian, and every firm with a chatbot on its website should learn it from Air Canada rather than from its own tribunal. The money is trivial. The principle is not: your bot speaks for your firm.

What the chatbot said

In November 2022, Jake Moffatt went to Air Canada's website after a death in his immediate family, looking for bereavement fares for flights between Vancouver and Toronto. The site's support chatbot told him, in part, that if he had already travelled he could submit the ticket for a reduced bereavement rate within 90 days by completing a refund application form. That statement was wrong. Air Canada's actual bereavement policy, on a linked page, said bereavement consideration does not apply after travel has been completed.

Relying on the chatbot, Mr Moffatt booked full-fare flights — $794.98 one way and $845.38 the other, in the tribunal's record — then applied for the partial refund the chatbot had described. Air Canada refused the retroactive discount. In February 2023 an airline representative acknowledged in writing that the chatbot had provided misleading words and said the issue had been noted for correction. The dispute still ended up before the British Columbia Civil Resolution Tribunal.

The argument Air Canada lost

Air Canada's defence included the submission that the chatbot was a separate legal entity responsible for its own actions. The tribunal called that position remarkable, in substance, and rejected it. The airline was responsible for all the information on its website, whether it came from a static page or a chatbot, and it had a duty to take reasonable care that its representations were accurate and not misleading. The tribunal found negligent misrepresentation: a duty of care, a false and misleading representation, reasonable reliance by the customer, and loss.

In Moffatt v. Air Canada, 2024 BCCRT 149, the tribunal ordered Air Canada to pay $812.02 CAD in total: $650.88 in damages, $36.14 in pre-judgment interest and $125 in tribunal fees, within 14 days. The decision was reported by CBC in February 2024. So when this headline says a tribunal charged it $812, read Canadian dollars and a small bill — attached to a principle with no size limit. The next invented policy need not be a bereavement fare. It can be your refund window, your cancellation term, your price, your promise about data.

Why a small firm is more exposed, not less

A small firm putting a chatbot in front of customers usually connects it to whatever documents are handy, launches it on every page, and checks it when someone complains. That is an unattended machine making representations in your name, at hours when nobody is watching. The attention you saved answering routine questions returns as dispute attention — except now it arrives with screenshots. Governance, in operator language, is deciding in advance what the bot may say, what it must never promise, and who reads what it said this week.

Three habits cover most of the risk. First, ground the bot in a short, owned set of answers for prices, policies and promises — and let it say it does not know, with a route to a person, outside that set. Second, forbid the categories that create liability: discounts, refunds, deadlines, eligibility and anything a customer could rely on to spend money. Draft those for human approval instead. Third, sample the transcripts weekly, fix the source answer rather than the single reply, and keep a record. Our governance checklist turns that into a one-page routine, and the pilot purgatory guide shows where unowned bots usually stall.

None of this argues against a customer chatbot. Routine questions are exactly what a machine should absorb, so your attention stays on the customer who actually needs judgment. It argues against an unsupervised one. A bot with clear limits and a weekly reader is symbiosis. A bot with your whole website as its script and nobody reading is a spokesperson you never briefed. The tribunal has now told us, in writing, whose words those are.

A one-afternoon audit any owner can run

Open your own site as a stranger would. Ask the bot ten questions a customer would actually ask before spending money: the price if it can be known, what is refundable, what is never refundable, the last day to cancel, what happens if you miss an appointment, whether a discount stacks, how long delivery or scheduling takes, who owns a mistake, what data you keep, and how to reach a person. Screenshot every answer. Then compare each one, line by line, against the policy you would defend in a dispute. Every mismatch is either a source document to fix or a question the bot should refuse and route. This test takes an afternoon because the list is short, and it is short on purpose: those ten answers are where reliance, and therefore liability, lives.

Pay special attention to confident completeness. The dangerous answer is rarely a wild invention; it is a half-true sentence with the limiting condition dropped, exactly as the bereavement answer dropped the condition that the linked page carried. Teach the bot to carry its conditions with it, to quote the limit in the same breath as the benefit, and to stop — visibly, politely, with a named route to a person — at the edge of what you have approved. A bot that says it does not know, and hands over well, loses you nothing. A bot that guesses your refund policy loses you the argument Air Canada lost, at whatever scale your firm trades.

Tell Ada where a customer could talk to a bot on your site today. She will flag who carries the risk and what it should never promise, in your free report at /audit/. Fourteen days and $812.02 CAD was Air Canada's version. Yours is preventable in an afternoon.

Sources: Moffatt v. Air Canada, 2024 BCCRT 149 (BC Civil Resolution Tribunal decision); CBC News, Air Canada found liable for chatbot's bad advice, February 2024. Amount stated in Canadian dollars as ordered.