In late 2022 a man in British Columbia needed to fly to a family funeral. He asked Air Canada’s website chatbot about bereavement fares. The bot told him he could book at full price and apply for the bereavement discount afterwards, within ninety days. That was not the airline’s policy, which required the discount to be approved before travel. He relied on the bot, flew, applied, and was refused.
He took it to the BC Civil Resolution Tribunal. Air Canada’s defence was remarkable: the chatbot, it argued, was a separate legal entity responsible for its own statements. The tribunal rejected that in Moffatt v. Air Canada (2024 BCCRT 149), found negligent misrepresentation, and awarded about $812 in total. The money is small. The ruling is not: a company is responsible for what its website says, and it does not matter whether a person or a model wrote the sentence. This is VH-001 in the Foundation’s hundred verified AI harms — the first one, on purpose.
The gate that would have caught it
The corpus writes it plainly: the company stands behind every statement its AI makes to a customer, and verifies policy answers against the source of truth.
Two halves, and the second is the practical one. A policy question has a right answer that lives in one place — the fare rules. The bot should read that place, not recall a plausible paraphrase of it, and where it cannot, it should say so and hand the customer to a person. The failure was not that a model produced a fluent wrong sentence; models do. The failure was that nothing stood between the sentence and a customer who would act on it with money and grief on the line.
Level 100 is where you start
You do not need a framework to see this. Level 100 asks three questions of one system you already run: what is it allowed to say or do, what does it check that against, and who finds out when it is wrong? For Air Canada the honest answers were: anything; nothing; the customer, three months later, at a tribunal.
The hour
This is the first door in the café — we’re using AI and nobody has written down what it’s allowed to do. Sixty minutes on paper: pick one assistant your organisation exposes to the public, list the five things customers ask it that involve a promise, find the source of truth for each, and write the line that says what happens when the bot and the source disagree. Then bring it on Thursday at 15:00 Pacific, where a host marks it.
The bot was confident and the policy was elsewhere. Sit down. What brought you in?