In February 2024, the British Columbia Civil Resolution Tribunal held Air Canada liable for negligent misrepresentation after its website support chatbot told a grieving customer he could claim a bereavement fare refund retroactively within 90 days — the opposite of the airline's actual policy. The tribunal characterized Air Canada's defence as suggesting the chatbot "is a separate legal entity that is responsible for its own actions," called that "a remarkable submission," and ordered CA$812.02 in damages, interest, and fees.
The system at the centre of Moffatt v. Air Canada (2024 BCCRT 149) was a support chatbot embedded in Air Canada's website, answering customer-service questions alongside the site's static policy pages — including a "Bereavement travel" page describing reduced fares for passengers travelling after a death in the immediate family. The tribunal record is notably thin on the system itself: the decision states that "Air Canada did not provide any information about the nature of its chatbot," and the tribunal fell back on describing it generically as "an automated system that provides information to a person using a website." Press coverage described it as an AI chatbot and quoted Air Canada's chief information officer framing the airline's AI investment as a customer-experience play — automating simple queries so humans could handle complex ones. Whatever the implementation, the deployment pattern was clear: the bot answered policy questions in Air Canada's name, on Air Canada's domain, with the same apparent authority as the airline's published policy.
On November 11, 2022, the day his grandmother died, Jake Moffatt asked the chatbot about bereavement fares before booking flights from Vancouver to Toronto. The bot replied — in text preserved as a screenshot in evidence — that a customer who has "already travelled" may "submit your ticket for a reduced bereavement rate ... within 90 days of the date your ticket was issued." That retroactive-claim window does not exist: the "Bereavement travel" page, which the chatbot's own answer hyperlinked, says the policy "does not apply to requests for bereavement consideration after travel has been completed." Relying on the bot, Moffatt booked two one-way flights totalling CA$1,630.36 and applied for the reduced rate on November 17 — well inside the promised 90 days. Air Canada refused the retroactive refund the chatbot had described, and months of correspondence failed to resolve it.
chatbot gives wrong policy answer that contradicts the page it links to; are we liable for what our chatbot tells customers; support bot invented a refund deadline that is not in our policy; customer relied on chatbot answer and is demanding a refund — the searches this case answers. Deeper in the incident: negligent misrepresentation ai chatbot case law; company legally bound by chatbot output; chatbot screenshot used as evidence in small claims; hallucinated 90 day refund window; can a chatbot be a separate legal entity; airline bereavement fare chatbot ruling.
The chatbot generated (or served) an answer about refund eligibility that was not grounded in the airline's canonical policy corpus, and the deployment had no mechanism to check the answer against that corpus before presenting it — even though the answer itself linked to the authoritative page it contradicted. Ungrounded output was rendered with exactly the same authority as documented policy, on the same domain, with no signal distinguishing the two. The tribunal made this the legal crux: Air Canada "did not take reasonable care to ensure its chatbot was accurate" and could not "explain why the webpage titled 'Bereavement travel' was inherently more trustworthy than its chatbot." The failure is the missing answer-to-source consistency check, not merely the single wrong sentence it let through.
Air Canada never published architecture details; the tribunal explicitly recorded that the airline "did not provide any information about the nature of its chatbot," so whether the bot was an LLM generating free text, an NLU intent system serving stale canned copy, or a hybrid is not publicly documented. What the record does establish is the mechanism class. There are two ways a support bot can answer a policy question: retrieve and quote the authoritative policy source, or synthesize an answer from whatever content, training data, or scripts it holds. The second path produces fluent text with no guarantee of fidelity to the current policy — and Air Canada's bot produced a specific, actionable term (a 90-day retroactive claim window via the "Ticket Refund Application form") that the policy did not contain, while hyperlinking the page that refuted it. That hyperlink is the diagnostic detail: the system knew which document was authoritative, yet nothing compared the answer's claims against that document before display. No grounding constraint, no entailment or consistency validation, no confidence gating, no visible provenance distinguishing generated prose from policy text. Any of those controls would have either corrected the answer or surfaced the contradiction; the absence of all of them is what turned one bad sentence into a binding misrepresentation.
The chatbot's wrong answer hyperlinked the correct policy page, so the contradiction was machine-detectable in principle and undetected in practice. Air Canada's litigation posture compounded the failure: it argued customers should have double-checked one part of its website against another — the tribunal rejected shifting that burden — and it asserted a tariff defence without filing the tariff text, drawing an adverse-inference warning from a tribunal that expected more from "a sophisticated litigant." An internal signal also went unheeded: a representative admitted the "misleading words" in February 2023, a year before the ruling, yet the dispute proceeded to a public decision.
On February 8, 2023, after Moffatt emailed the chatbot screenshot, an Air Canada representative admitted the bot had used "misleading words," pointed to the linked bereavement page, and said the issue had been noted "so it could update the chatbot" — the record does not document whether an update shipped. Air Canada also offered a $200 coupon, which Moffatt declined and the tribunal refused to count as a set-off. After the February 2024 ruling, press coverage (CMSWire, April 2024) reported the chatbot was no longer available on Air Canada's website; Air Canada did not publicly confirm the removal or explain it.
The ruling erased the legal distinction between a chatbot answer and published copy: "It makes no difference whether the information comes from a static page or a chatbot." A company is responsible for everything its site says, however the words are produced, and linking to the correct policy does not cure a wrong answer — the tribunal found it reasonable for a customer to trust the bot without cross-checking the very page it cited. The direct award was small (CA$812.02); the durable cost is the precedent, now cited across the legal industry as the default answer to "who is liable when the bot is wrong?" Teams shipping customer-facing bots should assume every generated sentence carries the same weight as their terms and conditions, because at least one tribunal says it does.
Court record
News
Blog
Blog
News
Permanently open — better fixes replace the top spot.
ROOT CAUSE ADDRESSED
The gap between what the bot asserts and what the canonical policy corpus actually says — ungrounded answer synthesis with no consistency check against the authoritative source the answer itself linked to.