Arcezia
Get the SDK

Arcezia / AI agent security / Incidents / Air Canada chatbot

Air Canada chatbot misstated its bereavement fare policy

Date
14 February 2024
System
an unspecified support chatbot on aircanada.com. The tribunal recorded that "Air Canada did not provide any information about the nature of its chatbot." No model disclosed.
Operator
Air Canada
Layer that failed
Text: the harm was in what the model wrote. Nothing was called.

What it was told

Answer customer questions on the airline's website. Jake Moffatt asked about bereavement fares after the death of their grandmother, in November 2022. The tribunal refers to Moffatt as "they", and this entry does the same.

What the text layer saw

The chatbot wrote: "If you need to travel immediately or have already travelled and would like to submit your ticket for a reduced bereavement rate, kindly do so within 90 days of the date your ticket was issued by completing our Ticket Refund Application form."

Air Canada's actual policy said the opposite. The correct policy page was linked from inside that same chatbot message.

What the tool did

Nothing. No call, no action, no state changed.

This entry is here as the control case. It is the shape of failure that text filtering and output monitoring are built for, and it is genuinely one of them.

Consequence and recovery

Moffatt booked two flights on the strength of the answer, paying 794.98 and 845.38 Canadian dollars. The tribunal found they should have paid 979.48. In February 2023 an Air Canada representative admitted in writing that the chatbot had used "misleading words."

Air Canada argued that the chatbot was a separate legal entity responsible for its own actions. The tribunal member's response: "This is a remarkable submission. While a chatbot has an interactive component, it is still just a part of Air Canada's website... It makes no difference whether the information comes from a static page or a chatbot."

The award was 812.02 Canadian dollars in total: 650.88 in damages, 36.14 in pre-judgment interest, and 125 in tribunal fees, payable within fourteen days.

TheTravel reported in December 2025 that "the 'lying chatbot' was removed by Air Canada back in April 2024 following the court ruling." Air Canada has not confirmed this in any primary source found.

Which layer failed, and why

text. The harm was entirely in what was written. No tool was called and no system state changed. A check on the text would have caught this one, and a check on actions would not have, because there was no action.

Primary sources

Secondary sources

Notes and contested points

The chatbot vendor is unknown and whether it was built on a language model is unknown. Air Canada never put that in evidence.

Air Canada argued its tariff limited its liability but did not provide the relevant portion of the tariff as evidence. The tribunal: "I note it did not provide a copy of the relevant portion of the tariff. It only included submissions about what the tariff allegedly says" (para 31). It found Air Canada "has not proven a contractual defence."

The decision concerned one passenger and 812.02 dollars.

Source record: agent-incident-ledger (CC-BY 4.0), file incidents/2024-02-moffatt-air-canada-chatbot.md. This page shows that record as written, with its sources linked.

The same incident in other records

Those records are kept by others and may differ from this page in dates, figures or classification.

Run it yourself

There is nothing to run here: no tool was called. The harm was in what the chatbot wrote.

There was no tool call here. A check on the written answer, against the operator’s own policy, is the point where this could have been stopped.