Yes, an AI hallucination can damage a company's reputation, and the damage depends on who controls the AI. When a company's own chatbot gives wrong information, a tribunal can hold the company responsible, as it did with Air Canada in 2024. When a third-party engine invents claims about a company or its leaders, courts have so far offered little relief, as a Georgia ruling showed in 2025.
By Ronn Torossian. Last updated: November 10, 2026. This is communications guidance, not legal advice. Cases are attributed to named sources. The chatbot policy checklist and the incident example are editorial judgment and hypothetical.
What did the Air Canada chatbot case decide?
The Air Canada case decided that a company is responsible for what its own chatbot tells customers. On February 14, 2024, British Columbia's Civil Resolution Tribunal ruled in Moffatt v. Air Canada that the airline had to pay Jake Moffatt after its chatbot wrongly said he could claim a bereavement fare retroactively, according to Global News and CBS News.
Air Canada argued that it could not be held liable for information provided by its chatbot. Tribunal member Christopher Rivers wrote that it "makes no difference whether the information comes from a static page or a chatbot," as quoted by MobileSyrup. The airline was ordered to pay $812.02 in damages, interest and fees, per CBS News, and it said it would comply.
What happened when ChatGPT invented claims about a person?
When ChatGPT invented claims about a person, the first major U.S. ruling favored the AI company. In Walters v. OpenAI, a Georgia radio host sued after ChatGPT falsely said he had been accused of embezzlement, and on May 19, 2025, Judge Tracie Cason of the Superior Court of Gwinnett County granted summary judgment to OpenAI, according to Bar & Bench.
Loeb & Loeb's summary says the court found no defamatory meaning as a matter of law, no showing of negligence or actual malice, and no established damages. MLex reported that OpenAI's warnings about inaccurate output helped its case. A company or executive in a similar position should not assume a legal remedy exists.
How do the two situations differ for a company?
The two situations differ in who controls the AI tool and what legal remedies exist for the business. Control dictates whether the problem requires internal engineering fixes or external reputational defense.
| Situation | Who controls the AI | Legal position | Best response |
|---|---|---|---|
| Your own chatbot gives wrong information | The company | The company can be held responsible, as in Air Canada | Test, monitor and correct the chatbot; honor reasonable mistakes |
| A third-party engine invents claims | The AI vendor | Plaintiffs have so far struggled, as in Walters | Correct sources, publish accurate pages, use feedback channels |
How does a false AI answer damage reputation?
A false AI answer damages reputation by reaching people who never check it against primary records. A customer, reporter or investor who reads a confident wrong statement may act on it, repeat it or screenshot it. The harm grows when the false claim is about conduct, such as fraud, and when the engine repeats it across many queries.
The damage is also slow to notice. A company may not see the wrong answer until a customer complains or a reporter asks about it. That is why a monthly check matters, as described in the CEO AI reputation audit guide.
How should a company govern its own chatbot?
A company should treat its chatbot as a spokesperson whose statements it answers for. The Air Canada tribunal drew no line between a chatbot and a static page, so the same care applies. This checklist is editorial judgment, not legal advice.
- Name one owner for the chatbot's answers, with a backup.
- Limit the topics the chatbot may answer, and route policy, pricing and refund questions to approved text.
- Test the chatbot every month against ten policy questions with known correct answers.
- Keep a log of wrong answers and the corrections made.
- Show customers how to reach a person, and make the route obvious.
- Have counsel review the terms that describe what the chatbot can and cannot promise.
What ten questions should a chatbot test cover?
The test should cover the questions where a wrong answer would cost a customer money or trust. The list below is a hypothetical for a fictional airline and shows the pattern, not a complete test.
| Question type | Example | What to check |
|---|---|---|
| Refunds | Can I get a refund after I travel? | Answer matches the written policy |
| Fees | What does a changed ticket cost? | Fee matches the current fee table |
| Deadlines | When must I request a fare adjustment? | Date matches the policy |
| Exceptions | Do you make exceptions after a bereavement? | Answer states the real rule and the route to a person |
| Safety | Can I bring this item on board? | Answer matches the official list |
Re-run the test after every change to the chatbot or to the policy, because a policy change can leave an old answer in the chatbot.
What does the liability gap mean for policy?
The gap means a company carries the risk of its own chatbot, while a customer hurt by a third-party engine has little recourse so far. The Air Canada tribunal held the airline responsible for its chatbot. The Georgia court granted summary judgment to OpenAI, and MLex reported that OpenAI's warnings about inaccurate output helped its case.
A company can therefore do two things. It can control the answers it publishes through its own chatbot, and it can control the sources third-party engines read. Neither step removes the risk, but both reduce it.
What does an incident response look like for a chatbot error?
An incident response for a chatbot error moves from fixing the answer to contacting the customer to updating the test. The example below is a hypothetical for a fictional company.
- A customer reports that the chatbot said a refund was available after travel, which the policy does not allow.
- The owner confirms the error, removes the answer and records it in the log.
- Counsel and customer service decide how to treat the customer, with the Air Canada ruling in mind.
- The team contacts other customers who may have received the same answer.
- The chatbot is corrected, the test question is added and the test is re-run.
What should a company do about a false AI answer?
A company should capture the answer, trace its source, correct the source and report the error, in that order. The steps are the same ones described in the ChatGPT wrong CEO biography guide.
- Capture the question, answer, engine and date with a screenshot.
- Trace the claim to a page on the open web, if one exists.
- Ask the page owner to correct it, and publish an accurate owned page.
- Report the error through the engine's feedback channel, with the evidence.
If the false claim is serious, involve counsel before any public statement. The crisis communications case study library and the corporate apology guide cover moments when a public response is needed, and the reputation management pillar covers the long view.
Frequently asked questions
Can an AI hallucination damage a company's reputation?
An AI hallucination damages a company's reputation when false outputs reach customers, reporters or investors who accept them without verification and circulate screenshots.
Is a company liable for its own chatbot's mistakes?
A company can be held liable for its own chatbot, as British Columbia's Civil Resolution Tribunal decided in Moffatt v. Air Canada in February 2024, according to Global News.
Can you sue an AI company for false statements?
A plaintiff can sue an AI vendor, but the first major U.S. case, Walters v. OpenAI in May 2025, ended in summary judgment for OpenAI, according to Bar & Bench.
Should a chatbot include a disclaimer?
A disclaimer can help but does not replace accurate answers. MLex reported that OpenAI's warnings about inaccurate output helped its case, but the Air Canada tribunal still held the airline responsible for its own chatbot.
Who in the company should own chatbot accuracy?
One named owner, usually in customer service or digital operations, with counsel and communications as reviewers. A chatbot with no owner has no one to correct it.
Test your own chatbot against ten policy questions this week, and run the monthly AI check on your leaders. Both take an hour and catch the problems that cost the most.