Air Canada's chatbot ruling: when a screenshot is all the evidence you need

What happened

In November 2022, Jake Moffatt visited Air Canada's website to book a flight following a bereavement ^[1]. He asked the airline's chatbot about bereavement fares. The chatbot told him he could book at the regular fare and apply for a bereavement discount afterwards by submitting the required documentation within 90 days.

That wasn't right. Air Canada's actual bereavement policy, on a separate page of the same website, required the request to be made before travel ^[2]. Moffatt booked the flight, travelled and then applied for the refund. Air Canada refused, pointing to its policy. Moffatt took the case to the Civil Resolution Tribunal of British Columbia.

The airline's defence was unusual. It argued that the chatbot was a separate legal entity, responsible for its own actions, and that Air Canada could not be held liable for information provided by what it called one of its "agents" ^[1]. The tribunal didn't accept that. It said Air Canada was suggesting the chatbot was "a separate legal entity that is responsible for its own actions", and ruled that the airline is responsible for all the information on its website, however it's presented.

Think of it like the butter in your fridge. If it's gone off, you don't get to say the fridge is a separate entity and not your problem. You bought the butter, you put it in the fridge, and if someone gets ill, it's on you. Air Canada put the chatbot on its website. What it said was theirs.

The evidence: a screenshot and an admission

Moffatt had a screenshot of the chatbot's advice. An Air Canada representative had already admitted in correspondence that the wording was misleading ^[1]. On the question of what the bereavement policy actually said, the airline offered no evidence to the contrary. The tribunal said it would assume such evidence wouldn't have helped Air Canada's case.

So the customer had a screenshot, the airline admitted the advice was wrong, and that was enough. The tribunal ordered Air Canada to pay Moffatt the difference between what he paid and what the bereavement fare would have cost, plus fees and expenses.

What the law says about telling people

Article 50 of the AI Act requires providers of AI systems that interact with natural persons to design them so that people are informed they are interacting with an AI, unless that is obvious from the circumstances ^[3]. For systems already on the market before 2 August 2025, this applies from 2 August 2026.

The rule doesn't say chatbots must carry a flashing warning. It says people have to know. A small notice that stays visible while the conversation happens is usually enough. The point is transparency, not alarm.

The Air Canada case happened before that rule came into force, but it shows why the rule exists. When people don't know they're talking to a machine, they tend to assume a human has checked the answer. They trust it more than they should.

The lesson for anyone running a chatbot

The lesson is short. You are responsible for what your chatbot says. If it gives wrong advice, you can't deflect liability by saying the AI did it. A chatbot on your website is as much yours as the text on your About page.

The evidence question is harder. In this case, the customer kept a screenshot and the airline didn't dispute what the chatbot had said. But what if the customer's screenshot shows one thing and you remember it differently? What if there's a dispute six months later and nobody has a record?

Screenshots are easy to take and hard to verify. A customer who feels wronged will keep one. If you don't keep your own record of what the chatbot said, the customer's version becomes the only version, and you're arguing from a weak position.

How a digital witness would work here

The thing to seal is each answer the chatbot gives, or each conversation when it ends. That means the question, the answer, the timestamp and, because Article 50 requires it, confirmation that the customer was told they were talking to an AI.

Sealing takes a fingerprint of the record, not the record itself, so the conversation stays in your own systems. You send the fingerprint to WitniumChain through the REST API, the SDK or the MCP server, and get a witness ID back. The first 2,000 witnesses a month are free.

If a customer later disputes what the chatbot said, you share the sealed record. They calculate the fingerprint themselves and compare it with the one that was sealed when the conversation happened. If it matches, the record hasn't changed. Change a single character and it won't match.

Anyone you share it with can check it, without having to trust you or anyone else. That's the point of a neutral witness.

What it doesn't do is prove the chatbot's answer was correct, or settle whether you're legally bound by what it said. It shows what was said and when, unchanged. The rest is for you, the customer and, if it comes to it, a tribunal to sort out.

How it would work

1. Customer asks the chatbot a question.
2. Chatbot replies. The system seals a record of the question, the answer and the time.
3. Six months later, the customer disputes the advice. You share the sealed record. They check the fingerprint and see it hasn't changed since the day of the conversation.

What this means for you

If you run a customer-facing chatbot, here are three questions worth asking this week:

1. Can we show what our chatbot said to a specific customer on a specific day, in a form that would hold up if questioned?
2. Do customers know they're talking to an AI, as Article 50 requires?
3. Who keeps the record: us, the customer, or nobody?

A screenshot is a customer's evidence. Decide now what yours will be.