Besides the Air Canada chatbot case, have Canadian courts dealt with AI mistakes?

Yes. In Zhang v. Chen (2024 BCSC 285), a BC court dealt with two non-existent cases invented by ChatGPT in a lawyer's court filing. In Ko v. Li (2025 ONSC 2766 and 2025 ONSC 2965), an Ontario court dealt with fake and mis-cited cases in a factum that the lawyer later said was prepared in part with ChatGPT.

Did the courts say using AI was the problem?

No. In Ko v. Li, the court wrote that the error was not using generative AI to help draft the document. The failure was signing and using it without making sure the cases were real and supported the argument. The court added that it is not the use of AI itself that is the concern.

What happened to the lawyers?

In Zhang v. Chen, the court refused special costs but ordered the lawyer to personally bear the extra costs her fake cases caused and to review her other files. In Ko v. Li, the court dismissed the first contempt proceeding on conditions. A separate contempt proceeding was started later in 2025 after the lawyer admitted that some of her earlier statements to the court were untrue.

These are lawyer cases. Do they apply to my business?

Not directly. They turn on lawyers' duties to the court, and how the law applies to your business is a question for a lawyer. The common thread is still useful: in both cases, the person who put their name on the work answered for it, not the AI tool. This is general information, not legal advice.

What is the simplest habit to avoid this kind of mistake?

Before you publish or send anything an AI tool helped write, check every fact, name, number and link against the original source. If you can't find the source, take the claim out.

After Air Canada: Canadian Cases Where AI Got It Wrong

Canadian courts have dealt with AI mistakes, in cases about lawyers. In Zhang v. Chen, a BC court dealt with a lawyer who filed two non-existent cases invented by ChatGPT (2024 BCSC 285), and in Ko v. Li, an Ontario court dealt with a factum containing fake and mis-cited cases (2025 ONSC 2766). In both cases, the person who used the AI’s work without checking it was held responsible, not the tool.

For a small business owner, that’s useful news: you can let AI draft your quotes, policies and customer replies and still keep wrong facts out of anything that goes out under your name, with the four-step source check at the end of this article.

This article is general information, not legal advice. For your own situation, speak to a lawyer.

We covered the Air Canada chatbot decision in Who is liable when your chatbot gets it wrong?. The two cases below involve lawyers, not small businesses, but the courts’ reasoning about checking AI output is plain enough for any owner to use.

What happened in Zhang v. Chen?

In a BC family law dispute, a lawyer’s notice of application relied on two cases that “were discovered subsequently to have been invented by ChatGPT”, according to the Supreme Court of British Columbia’s ruling of 20 February 2024. Opposing counsel could not find the cases, and a researcher they retained came to the view that the cases did not exist (2024 BCSC 285, paras. 9–11).

Justice Masuhara wrote that “Citing fake cases in court filings and other materials handed up to the court is an abuse of process and is tantamount to making a false statement to the court” (para. 29). He found the lawyer had no intent to deceive and refused the special costs the other side asked for (paras. 31–32). But he ordered that the “additional effort and expense” her fake cases caused “is to be borne personally” by her, and that she review all her files before that court for AI-generated citations (paras. 43–44).

One detail matters for anyone using a chatbot at work. The court called it “troubling” that the lawyer did not read or heed ChatGPT’s own warning that its output could be inaccurate (para. 36). The ruling closes: “generative AI is still no substitute for the professional expertise that the justice system requires of lawyers” (para. 46).

What happened in Ko v. Li?

At a motion in Toronto on 1 May 2025, Justice Myers of the Ontario Superior Court followed the links in a lawyer’s factum: one led to an unrelated commercial real estate case, and another led to a “404 Error - Page not found” (2025 ONSC 2766, paras. 5–6). Another case was cited for the opposite of what it decided (para. 11). In reasons released on 6 May 2025, he listed lawyers’ duties, including “to ensure human review of materials prepared by non-human technology such as generative artificial intelligence”, and ordered the lawyer to show cause why she should not be cited for contempt (2025 ONSC 2766, paras. 20 and 31).

In a second decision released on 20 May 2025, the court recorded that the factum had been prepared in part with ChatGPT and that the cases were AI hallucinations (2025 ONSC 2965, para. 23). Justice Myers was precise about where the fault lay: “The error was not delegating the factum or using generative AI to assist in drafting the factum.” It arose when the lawyer “signed, delivered, and used the factum without ensuring that the cases were authentic” (para. 59). He dismissed the show cause proceeding on two conditions: that she take the professional development courses she had committed to, and that she not bill her client for the research, the factum and the motion (para. 74).

The story didn’t end there. In September 2025, the lawyer wrote to the court admitting that some of her earlier statements about who prepared the factum were not true, and the court started a new contempt proceeding and referred carriage of it to the Attorney General of Ontario (2025 ONSC 6785, paras. 1–2 and 15). As of that December 2025 decision, no finding of contempt had been made, and the court appointed a lawyer as amicus curiae to assist it (2025 ONSC 6785, paras. 50 and 54).

How do the cases compare?

	Zhang v. Chen	Ko v. Li
Court	Supreme Court of British Columbia (2024 BCSC 285)	Ontario Superior Court of Justice (2025 ONSC 2766)
What went wrong	Two non-existent cases from ChatGPT in a notice of application	Fake and mis-cited cases in a factum, two relied on in court
How it was caught	Opposing counsel couldn’t find the cases	The judge followed the links during the hearing
Was AI itself the problem?	Court: AI is “no substitute” for professional expertise (para. 46)	Court: “it is not the use of AI itself that is the concern” (2025 ONSC 2965, para. 60)
Outcome	Extra costs borne personally by the lawyer; file review ordered	Show cause dismissed on conditions (2025 ONSC 2965); new contempt proceeding started later (2025 ONSC 6785)

Who is responsible when AI gets it wrong?

In these cases, the person who signed and used the work. In Ko v. Li, Justice Myers put it this way: “the proverbial buck stops with counsel” (2025 ONSC 2965, para. 21). In Zhang v. Chen, the lawyer herself told the court she had referred to the cases “without verifying the source of information” (2024 BCSC 285, para. 12). The Air Canada decision we covered earlier made a similar point about a company’s website. None of the three decisions treated the AI tool as the one to blame.

Does this apply to a small business that isn’t a law firm?

Not directly. These decisions turn on lawyers’ duties to the court, and we’re not lawyers; how any of this applies to your business is a question for one. What carries over is a habit. If an AI tool drafts your service page, a quote, a policy or a reply to a customer, and you send or publish it, you’re the one who put your name on it.

What’s a simple check before you publish AI-written work?

Call it the source check. Do the four steps in this order, before anything an AI tool helped write leaves your business. For a one-page notice or a customer reply, it’s two careful reads, not a project:

Check every fact against the original. Prices, hours, policies and dates should match your own documents, word for word where it matters.
Click every link. Both cases above were caught by someone following a reference.
Look up every name and source. If you can’t find it, take it out.
Read the tool’s own warning. Most AI tools say their output can be wrong. In Zhang v. Chen, the court noticed that warning was ignored.

A hypothetical example: a bakery in Guelph asks a chatbot to draft its December hours notice. The draft says the shop is open on Boxing Day. Step 1, checking the draft against the owner’s own schedule, catches it before the notice goes on the door, the website and Google.

What should a customer-facing assistant do?

The same principle applies when the AI talks to your customers directly. An assistant should cite the source of its answer and pass the question to a person when it doesn’t know. If you’re looking at one for your business, the five steps and their prices are on /express. Name and email are enough to get the price in writing.

Read next: AI Assistants for Small Business: A Practical Guide covers what an assistant should and shouldn’t answer. If you have an AI-drafted page or reply you’re unsure about, email [email protected] and we’ll reply with an honest next step.

Sources checked on 1 October 2026. This article is general information, not legal advice.

How we wrote this: drafted by the YourFirstAI team with AI assistance. We read each decision in full and checked every quote and paragraph number against it on 2026-10-01, then reviewed the article against our editorial rubric before publishing.