Full transcript
The published script of this episode, word for word. Numbers are written as digits. On-screen label for the generated scenes: Dramatisation. AI-generated. Sources are listed below the transcript.
0:00Can a company blame its own chatbot?
Can a company blame its own chatbot? One airline tried.
The Stop Button Files from AI Governance Chain replays real AI cases, and the control that stops them. File 01 starts in November 2022.
0:18What the chatbot said
A passenger's grandmother has just died. The passenger asks Air Canada's website chatbot about bereavement fares.
The chatbot says the reduced rate can be claimed within 90 days of the ticket being issued. So the passenger flies, then applies.
0:34The refusal and the ruling
The airline refuses. Its policy page said bereavement rates don't cover travel already completed.
In effect, said the tribunal, the airline treated its chatbot as a separate legal entity. A remarkable submission. Its finding: it makes no difference whether the information comes from a static page or a chatbot.
In February 2024, the airline was ordered to pay just over 812 Canadian dollars. Small money. Big principle.
1:02If your chatbot says it, you said it
On your website, if your chatbot says it, you said it.
1:06The control
The control I'd put in place: 1. Policy answers come only from the current policy text, with the source linked. 2. Promises, refunds and discounts go to a human or a hard rule. 3. Every answer is logged.
1:22Try it: Agent Jail
Watch rules stop a runaway agent in 30 seconds: run Agent Jail from AI Governance Chain, link below. Next file: the lawyers who trusted six cases that never existed.