Ninety days, according to the chatbot
On 11 November 2022 Jake Moffatt’s grandmother died in Ontario. That day he asked the chatbot on Air Canada’s website about bereavement fares. It told him that if he had already travelled he could apply for the reduced rate “within 90 days of the date your ticket was issued”. He booked at full price, flew, and applied. Air Canada refused. Its bereavement page said the policy did not cover requests made after travel.
Quick check
What did Air Canada argue before the tribunal?
Before British Columbia’s Civil Resolution Tribunal, Air Canada said it could not be held liable for information given by its agents, including a chatbot. The tribunal said that amounted to treating the chatbot as “a separate legal entity that is responsible for its own actions”, and called it “a remarkable submission”. The chatbot is still just part of Air Canada’s website, and the airline is responsible for all the information on it, whether it comes from a static page or a chatbot. On 14 February 2024 the tribunal ordered Air Canada to pay C$812.02.
The principle is the one most harms in this course turn on: responsibility lands where it would have landed if a member of staff had said the same thing.
Where the chain holds
Any deployed system sits in a chain: the model provider, the company that built a product on it, the organisation that deployed it, the professional who used the output, and the person affected. Details differ by country. Three patterns hold.
Professional duties stay with the professional. A lawyer who files invented cases is sanctioned as a lawyer, as in Mata v. Avianca. A doctor is judged by the medical standard of care.
What a business tells customers binds the business. The tribunal applied negligent misrepresentation, the ordinary rule that a business answers for careless statements a customer relies on.
Providers disclaim. A model provider’s terms usually supply output as is, exclude warranties and cap liability. Some offer business customers indemnities against specific copyright claims, which tells you which risks they have priced.
The law that is moving
The European Union’s revised Product Liability Directive, adopted in 2024, brings software, including AI systems, within product liability. It also eases the burden of proof where a claimant cannot be expected to explain a complex system. Elsewhere most claims still run through negligence, contract and consumer law.
One gap is open. When a general-purpose model produces a harmful output in a use nobody foresaw, through no obvious fault of the deployer, most countries do not yet say cleanly who pays. That does not mean responsibility floats free. Almost every harm so far has landed on someone with a duty they already had.
Three habits if your work touches others
Your name makes it yours. That is how the cases have gone.
Read the contract before the accuracy report. What is warranted, what is the cap, who indemnifies whom, and what happens when the underlying model changes?
Keep the record. Input, output, model version and date. A year later, a missing record does not protect you. It means the dispute is settled without your evidence.
Check your understanding
A clinic’s booking assistant tells a patient a treatment is covered by their insurer. It is not, and the patient faces a large bill. On the decided cases, where does responsibility most likely sit?
Pick the one you would defend. Your answer stays private.
Your place and first choices stay on this browser. Classroom credit is recorded separately.
Stuck on something?Ask Rukmini, your AI tutor