The chatbot, the screenshot and the tribunal

What happened

In 2024, a customer asked an airline's chatbot about bereavement fares and got the wrong answer ^[1]. He bought a ticket at full price, then applied for a partial refund on the grounds that the chatbot had told him he could ^[1]. The airline refused. The customer took it to the Civil Resolution Tribunal of British Columbia, and the tribunal ordered the airline to pay him ^[1].

The case is Moffatt v. Air Canada, decided on 14 February 2024 ^[1]. It's a short ruling, clearly written, and it's worth reading if you run a chatbot.

What the airline argued

The airline said it couldn't be held liable for information from its agents, including a chatbot ^[1]. In effect, the tribunal said, the airline was suggesting the chatbot was "a separate legal entity that is responsible for its own actions" ^[1]. The tribunal didn't accept that.

The airline is responsible for everything on its website ^[1]. It doesn't matter whether information comes from a static page or a chatbot. If the chatbot gives wrong information, that's the airline's problem.

The evidence

The customer had a screenshot of the chatbot's answer ^[1]. An airline representative had already admitted that the words in the screenshot were misleading ^[1]. On the fare itself, the airline offered no evidence to the contrary, so the tribunal assumed such evidence wouldn't have helped it ^[1].

That's the interesting part. The customer had his record. The airline didn't contradict it.

The lesson

The lesson is fairly simple. What an AI says on your website is yours. In a dispute, the side without its own records starts on the back foot.

A screenshot is the customer's evidence. What's yours?

What a digital witness does in this case

Here, the thing to seal is each answer as the chatbot sends it, or each conversation when it ends. Sealing takes a fingerprint of the answer, not the answer itself, so the conversation never leaves your systems. You send the fingerprint to WitniumChain through the REST API, the SDK or the MCP server, and get a witness ID back ^[3]. The first 2,000 witnesses a month are free ^[3].

If it's challenged later, anyone you share the record with can calculate the fingerprint again and compare it with the sealed one. If they match, nothing has changed. Change a single character and they won't match.

That gives you your own record of what the chatbot said, and when, in a way nobody can quietly rewrite afterwards. It doesn't show that the answer was correct, or settle whether the company is bound by it. But it means you're not relying on a customer's screenshot as the only record.

Say a customer claims the chatbot promised a refund that the terms don't allow. With a sealed record, you can show what the chatbot actually said. Without one, you're guessing.

How it would work

1. Your system sends a fingerprint of each chatbot answer to WitniumChain as it's sent.
2. You get a witness ID back and store it with the conversation.
3. If the answer is questioned, anyone you share it with can check that it's the one sent on the day.

Article 50 in the EU

Since 2 August 2026, people in the EU must as a rule be told when they're interacting with an AI system ^[2]. That includes chatbots. The AI Act calls it a transparency obligation ^[2].

It's a good rule. People should know when they're talking to a machine. But it doesn't settle the evidence question. When someone disputes what the chatbot said, who has the record?

What this means for you

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

1. Could we show what our chatbot said yesterday, if a customer turned up with a screenshot?
2. Where is that record kept, and who could change it?
3. If we sealed each answer or conversation as it happened, what would that cost us to build?

The answer to the third question might surprise you. The first 2,000 witnesses a month are free, and sealing a chatbot answer takes one API call ^[3].

What to watch

The transparency requirement in Article 50 of the AI Act has applied since 2 August 2026 ^[2]. Fines for breach can reach 15 million euro or 3 per cent of global annual turnover, whichever is higher ^[2]. For small and medium enterprises, the lower amount applies ^[2].

That's the rule on telling people they're talking to an AI. The evidence question is separate, and it's one you can answer before someone asks it.