You did not say it. Your AI did. You are still the one who answers for it. That is not hypothetical. Two companies already tried to argue otherwise, and lost.

A real AI redesign has to answer questions most org charts were never designed for. What responsibility changes when AI performs part of the work? What can a person override? And when the system gets something wrong, who owns the outcome?

The Air Canada Chatbot Case

In November 2022, a Canadian traveler named Jake Moffatt went to Air Canada’s website after his grandmother died, trying to book a last minute flight home for the funeral. Air Canada’s chatbot told him he could pay full price and apply for a bereavement discount afterward, within 90 days. He did exactly what it said.

When he filed the claim, Air Canada turned him down. Its actual policy did not allow a retroactive bereavement discount. The chatbot had given him the wrong answer.

Moffatt took the airline to the British Columbia Civil Resolution Tribunal. In February 2024, the tribunal ruled on the case. Air Canada’s defense was that the chatbot was, in the airline’s own words, “a separate legal entity responsible for its own actions.” The tribunal rejected that argument and held Air Canada responsible for the information provided through its website. The airline was ordered to compensate Moffatt.

A customer trusted the information his airline put in front of him. That is not unreasonable. It is the reason companies put chatbots on websites in the first place. The damages in this case were small. The exposure next time might not be.

The Same Defense, a Second Time

This was not an isolated case. In May 2026, the Oberlandesgericht Hamm, a regional appeals court in Germany, reached a similar conclusion in a different industry.

Aesthetify GmbH, a German aesthetic medicine clinic, operated a chatbot on its website to answer patient questions and book appointments. When asked, the chatbot told users that two doctors associated with the clinic held specific board certified specialist titles in plastic and aesthetic surgery. They did not hold those titles. Verbraucherzentrale Nordrhein-Westfalen, a German consumer protection organization, sued.

Aesthetify argued that the chatbot’s statements should not be treated as the company’s own statements. The court rejected that position, and concluded that deploying the system still mattered even though the chatbot had originally been trained on accurate information.

Designing for the Moment Things Go Wrong

A real AI redesign has to account for failure, not just success. What could go wrong? Where would the failure first appear? How would it be detected before it reaches a customer or decision maker? And who has the authority to intervene?

That last point matters. Responsibility without authority is not enough. If someone is expected to catch a problem but cannot pause the system, correct the output, or move the work into a manual process, accountability exists on paper but not in practice. None of that exists by default. It has to be designed.

AI Accountability Is Not Optional

The Moffatt ruling involved one customer, one wrong answer, and a relatively small claim. That is exactly what makes it useful. It shows what failure looks like at the low end of the risk scale. Many organizations deploying AI today are exposed to some version of the same problem, whether anyone has mapped it out yet or not.

The fix is not to slow AI down. It is to put a name on who owns the outcome before something goes wrong, not once someone else is already asking who’s responsible.

When you are presented with a new AI solution or agentic process, ask two questions: if the AI system fails, who owns the outcome and who is accountable? And if it fails, who is empowered to step in, and are they equipped to respond?