Blog · 2026-10-15 · baryreview editors

Who is liable when your AI-written content is wrong?

First, the disclaimer

This is not legal advice. We are editors, not lawyers. The law on liability for AI generated content is new, varies by country and state, and is being worked out case by case. If you have a specific problem, talk to a lawyer who knows your jurisdiction. What we can do here is lay out the handful of cases that get cited most often, describe what the sources actually say, and draw the practical lesson for anyone publishing AI-assisted content.

That lesson is short. In every case so far, nobody has been able to point at the AI and walk away.

The principle: the publisher owns what it publishes

Before AI, this wasn’t controversial. If your website said your product was waterproof and it wasn’t, that was your statement. If your blog post said a competitor was under investigation and it wasn’t, that was your statement too. The person or tool that drafted the sentence never mattered to the reader, the customer, or the court. What mattered was whose name was on the page.

Generative AI tempted some companies to test whether that had changed. The first notable test went badly for the company.

Moffatt v. Air Canada: “it is still just a part of Air Canada’s website”

In November 2022, Jake Moffatt used the chatbot on Air Canada’s website to ask about bereavement fares. The chatbot told him he could buy a ticket and apply for the bereavement discount afterward. Another page on the same website contradicted that. Air Canada refused the retroactive refund, and Moffatt took the airline to British Columbia’s Civil Resolution Tribunal.

Air Canada’s defence is the part everyone remembers. According to the tribunal’s decision, Moffatt v. Air Canada, 2024 BCCRT 149, as summarized by McCarthy Tétrault, the airline argued it could not be held responsible for information provided by the chatbot. The tribunal member wrote: “In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions.” He rejected that. “While a chatbot has an interactive component, it is still just a part of Air Canada’s website. It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.”

The tribunal found negligent misrepresentation: “I find Air Canada did not take reasonable care to ensure its chatbot was accurate.” The decision was released on February 14, 2024. The amount was small, C$650.88 in damages and C$812.02 in total with interest and fees, per Law360 Canada. The principle is not small. A business publishes what its tools say, and it has a duty to take reasonable care that it is accurate.

Note what the tribunal did not require. It did not require Air Canada to have known the chatbot was wrong. The standard was reasonable care. The airline had a correct page on the same site; the problem was that nobody checked the chatbot against it.

Wolf River Electric v. Google: an ongoing case about an AI Overview

The second case runs the other way: a business claims it was harmed by someone else’s AI output. It is unresolved, and we describe it only as the sources do.

Wolf River Electric is a solar installer in Isanti, Minnesota. In March 2025 it sued Google in Minnesota state court. According to the complaint as reported by the Star Tribune and Minnesota Lawyer, a Google AI Overview shown for searches about the company stated that it was facing a lawsuit from the Minnesota Attorney General for deceptive sales practices. The company says that is false: the Attorney General had sued other solar-lending companies, not Wolf River, and none of the four sources the AI Overview cited supports the claim. The complaint alleges that a customer raised the claim in March 2025 and terminated a contract worth about $150,000, and that others followed. The company seeks damages it puts between $110 million and $210 million, on claims of defamation, defamation per se, defamation by implication, and violation of the Minnesota Deceptive Trade Practices Act.

Google removed the case to federal court in June 2025 and filed a response denying the allegations. In January 2026 the federal court remanded the case to Minnesota state court, where, according to a February 2026 client alert by Quinn Emanuel, it remains in its pre-trial phase. Nothing has been decided on the merits. Google has not been found liable for anything, and may never be. We cite the case because of what it illustrates, not because of how it will come out.

Two things in that client alert matter for ordinary publishers. First, the firm notes that an earlier AI defamation case, Walters v. OpenAI, was decided for OpenAI partly because the user had been warned that outputs could be inaccurate and could verify them. Context and disclaimers mattered. Second, the alert warns that people who republish AI-generated falsehoods may face liability themselves, and recommends verifying AI-generated factual claims before publishing them. In other words: whatever happens to Google, if you copy an AI’s false claim about a real company into your own blog post, the question of who is liable gets much simpler, and the answer is you.

The professional-services version: refunds, not lawsuits

Most businesses that publish wrong AI-assisted content will never see a courtroom. They’ll see a client asking for their money back.

The clearest example is Deloitte Australia. In 2025 it delivered a 237-page report, under a contract worth about AU$440,000, to the Department of Employment and Workplace Relations. A University of Sydney researcher found that it cited academic papers that do not exist and quoted a federal court judgment that said no such thing. The revised version disclosed that Azure OpenAI had been used in preparing it. Deloitte agreed to repay the final installment of its fee, and said the “matter has been resolved directly with the client,” according to AP. The department said the report’s findings and recommendations were unchanged.

Two months later, Fortune reported that a separate Deloitte report for the government of Newfoundland and Labrador, costing just under C$1.6 million, contained citations to papers that could not be found and listed researchers on papers they said they had not worked on. Deloitte said it “firmly stands behind the recommendations,” that AI was not used to write the report, and that it was “selectively used to support a small number of research citations.” It said it was correcting the citations. No refund had been reported as of that article.

An October 2026 piece in The Conversation, by two business-school academics, puts the Australian refund alongside the Air Canada decision and the court sanctions against lawyers who filed AI-fabricated citations, and reaches the conclusion we’d reach: “AI may change how professional work gets done, but it does not change where the buck stops.”

For an agency, a consultant, or a freelancer, this is the version of liability that actually arrives. Not a lawsuit. A client who found a fake citation, lost trust in the whole deliverable, and wants a refund or a discount. The contract may not even mention AI. It doesn’t need to; it says you were paid to produce accurate work.

What this means for a business blog

Put the three cases together:

• A tribunal held a company to a reasonable-care standard for what its website said, and treated an AI tool as part of the website.
• A company is suing over an AI summary that it says falsely accused it of being sued, and a law firm’s commentary warns that republishing such a claim creates your own exposure.
• A consultancy refunded part of a fee because a report contained fabricated sources, even though the client accepted the report’s conclusions.

The common thread is that the reader, client, or court looks at the name on the page. Nobody asks what model you used. If the page is wrong, it’s your wrong page. That is true for a landing page claiming a feature your product doesn’t have, a blog post citing a study that doesn’t exist, a comparison article that says something false about a competitor, and a help article that promises a refund your policy doesn’t allow. Items 1, 2, and 7 on our list of what we check exist because these are the paragraphs that cost money.

What to do before publishing AI-assisted content

None of this means stop using AI. It means publish the way a careful company has always published: someone checks, someone signs, and there’s a record.

A fifth item for anyone publishing on behalf of clients: put AI use and verification responsibility in the contract. Say what tools you use, what you check, and who is accountable for accuracy. Deloitte’s Australian report disclosed its AI use only in the revised edition; the disclosure would have been cheaper on day one.

Where an outside review fits

A single flat-fee editorial review doesn’t change your legal position, and we won’t pretend it does. What it changes is the odds. Our editors read an AI-written article the way an informed reader and an opposing lawyer would: every fact, every citation, every claim about your product or someone else’s. The corrections come back as instructions you paste into your own AI tool, so the page gets fixed before it is the page you are liable for. See /how-it-works for the process and /sample-review for what the corrections look like.

Practical takeaway

Assume you are liable for every sentence on your site, because so far that is how it has gone. Verify claims against sources you opened yourself, keep the drafts and sources, have a named person sign off, and fix reported errors fast. The AI didn’t publish the page. You did.