Air Canada: a fare difference and a “remarkable submission”
In November 2022, after their grandmother died, Jake Moffatt asked the chatbot on Air Canada’s website about bereavement fares. The bot said the reduced fare could be claimed after the fact, by requesting a refund within 90 days of the ticket being issued. The airline’s actual policy did not allow that. The dispute over the difference went to the Civil Resolution Tribunal, British Columbia’s small-claims tribunal.
As the tribunal described it, Air Canada in effect argued that the chatbot was a separate legal entity responsible for its own actions. Tribunal Member Christopher Rivers called this “a remarkable submission”. In decision 2024 BCCRT 149 of 14 February 2024 he wrote: “It should be obvious to Air Canada that it is responsible for all the information on its website” (para. 27). The same paragraph says it makes no difference whether the information comes from a static page or a chatbot. The next paragraph draws the conclusion: the airline did not take reasonable care to make sure its chatbot was accurate.
The arithmetic is short. Moffatt paid C$1,630.36; the bereavement fare would have been C$979.48. The difference of C$650.88 is the loss. On top of that, C$36.14 in interest and C$125 in tribunal fees. Total: C$812.02.
Germany, 2026: the bot gave the doctors the wrong specialism
An aesthetic medicine clinic in Recklinghausen (Aesthetify GmbH, whose doctors are known as “Dr. Rick & Dr. Nick”) put an AI chatbot on its website. Asked about qualifications, it said both doctors were specialists in plastic and aesthetic surgery; in other conversations, specialists in aesthetic medicine or aesthetic treatments. They hold no such specialist titles.
The consumer protection centre of North Rhine-Westphalia (Verbraucherzentrale NRW) went to court to stop those claims. On 12 May 2026 the Higher Regional Court of Hamm (case I-4 UKl 3/25) found the bot’s answers to be a misleading commercial practice under the Unfair Competition Act and barred the clinic from using them.
The court looked separately at whose words these were. The clinic could not shift the blame onto someone else. The judgment says its AI chatbot “is not a third party” in the sense of the rules on liability for other people’s infringements (translation ours). To the court, the bot was simply a technical means the clinic used to talk to prospective patients.
The second finding is closer to everyday practice. In the court’s view, a conscientious business would have seen it coming: the question about the doctors’ qualifications was an obvious one, and the bot could make up a wrong answer to it. So that answer should have been checked during setup.
Whether the ruling is final is less clear. The court allowed an appeal on points of law (Revision) to the Federal Court of Justice. Verbraucherzentrale NRW describes the ruling on its website as final. Both can be true if nobody filed that appeal, but we found no Federal Court of Justice case number.
New York: a bot that advised breaking the law
In autumn 2023 Mayor Eric Adams’s administration released MyCity, a chatbot for business owners built on Microsoft Azure AI. On 29 March 2024 The Markup published an investigation by Colin Lecher into its answers. One of them: “Yes, you can take a cut of your worker’s tips.”
It was not the only answer of that kind. The bot said landlords did not have to accept Section 8 housing vouchers, although source-of-income discrimination is illegal in New York City. It said a restaurant could go cash-free, although a 2020 city law requires businesses to accept cash. The bot’s page warned that it could occasionally give wrong, harmful or biased answers. A spokesperson for the city’s Office of Technology and Innovation (OTI) told the reporters it was a pilot and would improve. The bot stayed online.
In late January 2026, at a press conference on the city’s $12 billion budget gap, the new mayor, Zohran Mamdani, called the bot “functionally unusable” and put shutting it down on his list of savings. By his account, it had cost the administration around half a million dollars. According to The Markup and THE CITY, building the bot’s foundations alone cost nearly $600,000. By 4 February the bot was offline, and its page said the beta test had ended. From the investigation to the shutdown: 22 months.
There was no court case here. What the story does show is that errors found by journalists did not, by themselves, take the bot down.
What these cases are not
Moffatt v. Air Canada is usually retold as “court orders airline to pay”. More precisely:
- It is a small-claims tribunal, not a court. Its decision is binding within British Columbia’s small-claims system. For other provinces and countries it is not a precedent, although it is widely cited.
- The legal ground is negligent misrepresentation; contract has nothing to do with it. The tribunal asked whether the company took enough care over the accuracy of the information on its website.
- The sum is small: the loss, with interest and fees. There is no fine.
- In Hamm the claimant was a consumer association, not a patient, and it sought a ban on the false titles, not compensation for patients. The Federal Court of Justice may still have the last word.
The two decisions have one thing in common: neither the tribunal nor the court would separate the bot from the company that put it there.
What Uzbek rules already require
We found no Uzbek case law on chatbot answers. Rules that bear on them directly sit in two documents from 2026.
The first is Law ZRU-1115 of 21 January 2026. It added Article 7¹ to the Law “On Informatization”: legally significant decisions affecting a person’s rights and freedoms may not rest solely on the conclusions of AI-based systems. The final text has no mandatory labelling of AI content. The draft did have it: according to Gazeta.uz’s report on the Senate approving the law on 1 November 2025, the bill required labelling of all information resources created with AI. That did not make it into the signed text.
The second is the Ethical Rules for the Creation, Deployment and Use of AI-Based Solutions. The Ministry of Digital Technologies approved them by order No. 284-mh of 24 February 2026, the Ministry of Justice registered them on 14 March under No. 3787, and they came into force on 17 June. Lex.uz carries only the Uzbek text. Three items matter for a website bot:
- item 6: participants should not treat AI decisions as final;
- item 10: users must be told that AI decisions are not always correct and that their results can be used for advisory or reference purposes;
- item 11: decisions made with AI must be under human control and reviewed by a human, and the final decision must be taken by a human.
Applied to a bot on a website, that comes out as: the customer has been told an AI is answering and may be wrong, and decisions about money and exceptions to the rules are made by a member of staff.
Six limits to write down before launch
All three stories come down to one thing: a bot has to be set up like an employee whose words the company will later have to stand behind. Here is what to put in the bot’s brief before it answers its first customer.
- Where answers come from. Only from a document the owner wrote: price list, terms, common questions. If something is not in it, the bot does not know it and says it will check.
- Money. Prices only from the price list. No discounts, instalments, refunds or compensation unless the document provides for them. This is exactly what Moffatt was about. If the bot also sells, Uzbekistan has rules for the offer: what it must show before payment is covered in our piece on Telegram shops and ZRU-792.
- Qualifications and guarantees. Licences, titles, deadlines and outcomes are given only in the document’s own wording. List the questions customers ask your business most often and check the bot’s answer to each one. That is what the Hamm court expected of the clinic.
- An “AI is answering” note. In its first message the bot says what it is, and it does not deny it when asked directly.
- Handover to a person. A list of topics where the bot stops and calls a member of staff: complaints, refunds, exceptions to the rules, health, contracts. And a way for that person to take over the conversation.
- A log. Conversations are kept, and someone reads them. Otherwise you find out what the bot promised from a customer’s complaint. The public learned about New York’s bot from a journalists’ investigation.
All three stories are about website bots, but on our reading their logic is not tied to the channel: a Telegram bot and a WhatsApp bot also answer customers on the company’s behalf. If a contractor builds the bot for you (we, for example, build chatbots for business in Telegram and WhatsApp), put this list straight into the specification.
Valli, our AI auto-responder for Telegram Business and website widgets, is currently in pilot. It is exactly this case: a bot writing to customers on the owner’s behalf. Some of the limits above are written into the rules the model receives with every message. It gives prices and deadlines only from the owner’s profile and catalogue, holds the price when a customer haggles, and when the profile has no answer it says it will check with the owner. While the profile is empty, it is not allowed to name prices, deadlines or services at all. It does not confirm meetings itself; the owner does. In its first message Valli introduces itself as the owner’s AI assistant, and it does not deny being a bot when asked directly.
A model can break rules too, so some of the work is done by the server without it. The model never writes out payment details: the server sends them to the customer in a separate message, word for word from the settings. If the owner replies in a Telegram chat, Valli stays silent in that chat for two hours, and a button mutes it in a chat for 24 hours. It keeps conversation context for 7 days, or 30 with the memory skill; the conversation itself stays in the owner’s Telegram.
Valli has no separate rule on refunds or qualifications. They fall under the general rule “not in the profile, check with the owner”, so everything depends on what the owner put in the profile. That has to be tested before launch, with your own questions.
What we don’t know
This is not legal advice. The British Columbia tribunal’s decision is binding only within that province’s small-claims system. In the Hamm case the court allowed an appeal on points of law, while Verbraucherzentrale NRW calls the ruling final; we have not established whether an appeal was filed. We found no Uzbek case law on chatbot answers, and it is not yet clear how Ethical Rules No. 3787 will be applied to private businesses. We found no official Russian or English text of the rules; the translation of the items is ours.
For an assessment of your particular bot and your terms with customers, talk to a lawyer.
If you already run a bot or are only planning one, describe the process in the quiz on the main page. Within 48 hours, free and with no intro call, we will map which questions the bot can handle on its own and which it must hand over to a person.
