A tribunal held Air Canada to what its chatbot told a customer about bereavement fares

The failure in Moffatt was two sources of truth on one website. The bereavement policy page did not allow retroactive claims. The chatbot told the customer he could book now and claim the discount retroactively within 90 days ^[1][2]. The tribunal did not weigh one against the other. It held that Air Canada was responsible for all the information on its website, static page or chatbot alike. The tribunal also found the airline owed the customer a duty to take reasonable care that its representations were accurate ^[4]. Air Canada was ordered to honour the fare and pay damages ^[5].

The airline's defence was that the chatbot was "a separate legal entity that is responsible for its own actions" ^[3]. The tribunal rejected it ^[4]. Applied to anything you ship, the holding says that policy wording your assistant can emit is policy wording your company has published ^[3]. That is an argument for retrieving and quoting the canonical text, and for refusing outside the retrieved scope, because both shrink the set of sentences you are on the hook for.

The vendor contract puts a ceiling on the vendor's share of this. Anthropic's consumer terms cap total liability at the greater of the amount you paid in the six months before the claim and $100, and exclude indirect and consequential damages ^[9]. OpenAI's provide the service "as is", disclaim warranties and exclude the same categories ^[10]. If your payments to the vendor over those six months came to less than $100, the ceiling is $100 ^[1]. Those clauses bound what a user can collect from the vendor. A customer in a Moffatt-shaped dispute is not suing the vendor; he is suing the business whose site answered him, and that claim lives in a different contract with different parties ^[2]. The dev.to account gives both sets of terms in summary and does not quote the clause text ^[15].

Garcia v. Character Technologies is the case that reaches further into product liability. It was filed in October 2024 by a mother after the death of her fourteen-year-old son, and named the chatbot maker, its founders and Google ^[6]. In an early 2025 ruling, a federal judge in the Middle District of Florida denied the companies' motion to dismiss, rejected a First Amendment argument that a chatbot's outputs were protected speech, and treated the app as a "product" for product-liability purposes, allowing wrongful-death, negligence and product-liability claims to proceed ^[7]. That ruling was about whether the case could go forward, and the matter later settled, so it produced no finding of liability ^[8].

Moffatt is a small-money consumer case from one province's civil resolution tribunal ^[13][1]. The dev.to piece is direct that no single law says who pays when an AI gets it wrong, that the answer depends on where you live and who you sue, and that claims are being fitted into contract, consumer protection, negligence and product liability ^[11]. Outside British Columbia, deployer-pays is only a direction. In my view it is still the assumption to build against. The corrective work is all on the deployer's side. On the vendor side there is the gap the piece describes between marketing that calls these tools ready for real work and contracts that call them experimental, unwarranted and use at your own risk ^[12].