Air Canada chatbot

Air Canada ran a support chatbot on its own website to answer customers' questions about its policies. Its answers appeared with the same authority as the airline's published policy pages. The record does not name who built the chatbot.

What happened

A customer arranging travel after a death in the family asked the chatbot about bereavement fares. The chatbot said they could apply for the reduced fare retroactively, after booking. That was not Air Canada's policy.

The customer relied on the answer, booked the flight, and then submitted the claim. Air Canada's human staff refused it, pointing to the actual policy the chatbot had misstated.

The ruling

The customer took the dispute to British Columbia's Civil Resolution Tribunal. The case is Moffatt v. Air Canada, 2024 BCCRT 149, decided on February 14, 2024. The tribunal found the airline liable for negligent misrepresentation. It awarded 650.88 Canadian dollars in damages, plus fees.

The damages were small. The case matters for the precedent it sets. The ruling is decided, with published reasons, so this case states its holding as a finding.

What each side said

Air Canada argued that the chatbot was a separate legal entity responsible for its own actions. On that argument, the airline would not be liable for what the chatbot said.

The tribunal rejected it. It held that the airline is responsible for all the information on its website, whether it comes from a static page or a chatbot. A customer has no way to know, and no obligation to work out, which of the two sources to trust.

The duty the ruling sets

An organization must take reasonable care that what its chatbot tells customers is accurate. It owes the same care for its published pages. The chatbot is a tool the organization deploys, not an entity that answers for itself.

The case file reads that duty this way. Every system that writes its own answers will sometimes state something false. A made-up policy is a known failure, not a freak event. So the duty is not to make the chatbot never wrong. It is to build an accuracy check on what the chatbot says, and to own its answers. A disclaimer that the AI speaks only for itself does not move that responsibility.

Where the chatbot and the staff disagreed

The chatbot said yes, and the staff said no. The customer was left holding the difference.

Staff were the point where Air Canada could have honored or corrected the chatbot's answer before it became a harm. Instead, the gap between the chatbot's promise and the staff's refusal was the injury. The record documents no process that kept the chatbot's answers in line with the actual policy before the ruling.

After the ruling

The record reports that Air Canada removed the chatbot from its website after the ruling. It does not say what, if anything, replaced it.

What this network is drawn from

This network follows the accountability pattern in the case file. It is not a reconstruction of Air Canada's actual system. It shows the chatbot, the staff who handle customer claims, the ownership of the chatbot's answers, the policy records, and the published policy pages. No customer outcome is computed on it.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network where a mistake made by one part can be passed on to the other.

Explore (No Targets) sets no targets. Under Service Targets Only, the targets can be met within this case's budget of 8 units. The cheapest way costs 2 units and uses one tool, Mark AI-written records. Other ways pair two tools, such as Review on schedule and Escalate checks for 4 units.

Under Service and Safety Targets and All Governance Targets, the targets are not fully addressable with the available tools. One failure pathway stays open whatever you choose: Staff set what the chatbot may state. In this example, a mistake in what staff set for the chatbot would be repeated to customers by the chatbot. None of the tools offered here acts on it.

The closest you can come costs 7 units: Gate record entries, Mark AI-written records, and Escalate checks. It closes every other failure pathway and meets the service targets. Adding Check with a second model, the tool that adds the accuracy check, would cost 3 more units, over the budget, and would still leave that pathway open. Using every tool at its strongest setting, at any cost, still leaves that pathway open, and the added checks leave the chatbot too little useful work.