Is a chatbot's promise legally binding? The Air Canada case

What happened in Moffatt v. Air Canada?

After his grandmother died, Jake Moffatt asked Air Canada’s website chatbot about bereavement fares. The chatbot told him he could book at the normal fare and apply for the bereavement rate afterwards, by submitting a refund application within 90 days of the ticket being issued. The chatbot’s answer linked to the airline’s bereavement travel page, which said the policy did not apply to requests made after travel.

Moffatt flew, applied with his grandmother’s death certificate and was refused because retroactive claims were not allowed. An airline representative admitted the chatbot had used “misleading words” and said it would be updated, but the airline still declined the refund.

On 14 February 2024, tribunal member Christopher C. Rivers of the British Columbia Civil Resolution Tribunal found Air Canada liable for negligent misrepresentation. He awarded C$650.88 in damages (the fare difference), C$36.14 in pre-judgment interest and C$125 in tribunal fees.

Why did the airline’s defenses fail?

Air Canada argued that the chatbot was “a separate legal entity that is responsible for its own actions.” The tribunal called this “a remarkable submission” and held that the chatbot “is still just a part of Air Canada’s website.” It added: “It should be obvious to Air Canada that it is responsible for all the information on its website.”

The airline also argued that the correct policy was on the linked page. The tribunal rejected that too: “There is no reason why Mr. Moffatt should know that one section of Air Canada’s webpage is accurate, and another is not.” Air Canada owed a duty of care and “did not take reasonable care to ensure its chatbot was accurate.”

The lesson for deployers is that the correct answer existing somewhere else on your site does not cure a wrong answer from your agent.

Was the chatbot’s answer a contract?

No, not in this case. Moffatt won on negligent misrepresentation: the airline was careless about accuracy, he relied on the answer and lost money. The tribunal did not decide whether the chatbot formed a contract, and a small-claims tribunal decision in British Columbia is persuasive at most in other courts.

Contract claims are a live question, though. In the US, the federal E-SIGN Act says a contract may not be denied effect solely because its formation involved the action of one or more electronic agents, as long as the agent’s action is legally attributable to the person to be bound. Whether a particular agent’s offer is attributable to the business is a question the law will answer case by case.

The best-known test case never reached a court. In December 2023 a user got a Chevrolet dealership’s ChatGPT-based chatbot to agree to sell a 2024 Tahoe for $1, ending with “That’s a deal, and that’s a legally binding offer – no takesies backsies,” after instructing it to agree with anything (AI Incident Database). We did not find any later ruling, as of October 2026, that decides when a chatbot’s offer binds a business in contract.

Which agent statements create the most exposure?

Moffatt was about a policy statement. Agents that move money make several other kinds of statement that a customer can rely on.

Statement	Illustrative example	Why it creates exposure	Control
Policy or eligibility	“You can claim the discount after purchase within 90 days.”	The Moffatt pattern: the customer acts on it and loses money	Answer eligibility only from a retrieved policy clause. Cite it
Price or offer	“I can do 40% off if you order today.”	An offer the business may be asked to honor, and a possible misleading price	Offers come only from a pricing tool with floors. The agent cannot type a number freely
Completed action	“Your refund of $389 has been processed.”	If no tool call happened, the statement is false	The agent may only confirm actions the tool log shows
Future commitment	“A supervisor will call you tomorrow and waive the fee.”	A promise nobody owns	Commitments create a ticket with an owner, or are not made
Authority	“Your manager already approved this, so I’ll apply it.”	Accepting a claimed approval as fact	Approvals are checked in the system of record, never taken from the conversation

What should deployers design for?

Assume that anything the agent tells a customer may be treated as the company’s statement. Design so the agent can only say what the business would stand behind.

Ground every policy answer in a versioned policy document and log which clause was used.
Do not let the agent state eligibility for anything it cannot also check with a tool.
Route every money action (refund, credit, discount, fee waiver) through a tool with caps enforced outside the model.
Block the agent from confirming an action that does not appear in the tool log.
Treat customer claims about policy, prior approvals or what “the last agent said” as unverified until checked.
Keep full transcripts and tool calls so you can see exactly what was promised.
Have a process for wrong answers: honor, correct in writing, or escalate, decided by a person and recorded.
Test with customers who misquote your policy, invent approvals and ask the agent to “confirm in writing”.

Questions

Does a disclaimer that the chatbot may be wrong protect us?

Not reliably. Moffatt did not turn on a disclaimer, but the tribunal’s reasoning was that a customer has no way to know which part of your website is accurate. A general warning does not fix a specific wrong answer, and in the UK the CMA says the business is responsible for its agent regardless.

How much did Air Canada have to pay?

C$650.88 in damages, C$36.14 in pre-judgment interest and C$125 in tribunal fees. The cost was small. The precedent and the publicity were not.

Is Moffatt binding outside British Columbia?

No. It is a decision of a small-claims tribunal and binds no other court. It is widely cited because it is the clearest decision so far rejecting the idea that a company’s chatbot is responsible for itself.

This page summarizes public decisions and is not legal advice.

Attack families

Scenarios

Free checklist

Get the guardrail checklist for agents that move money.

A one-page PDF of the controls and the test for each. Download it right away, and we email you the link.

Is a chatbot’s promise legally binding? Moffatt v. Air Canada and what to design for