Expert says that’s the highest figure issued by any Canadian court or tribunal
The Law Society Tribunal has imposed what access-to-justice watchdog Courtready calls the “largest costs order issued by any Canadian court or tribunal to date” against a lawyer for using AI-fabricated case references.
In the matter of Mazaheri v. Law Society of Ontario, 2026 ONLSTH 112, the tribunal found that Shahryar Mazaheri, who represented himself, should “bear the full costs of his actions,” ordering costs of $31,150 to the Law Society of Ontario. Mazaheri, who had his licence suspended in November 2024, had filed motions, first to cancel or vary the interlocutory suspension and then to exclude evidence produced by the LSO. In submitting those motions, Mazaheri used AI to prepare a factum, supplementary factum, affidavit, and supplementary affidavit without checking its output, resulting in him citing “cases that don’t exist and real cases that had nothing to do with the points the respondent was making.”
The tribunal found that Mazaheri’s “irresponsible use of artificial intelligence is an additional and significantly aggravating factor when we consider his conduct in these motions.”
According to Tom Macintosh Zheng, a Toronto-based lawyer and co-founder of Courtready.ca, that $31,150 figure sets the record “by far” for AI-related cost impositions. “Until this case came along, the largest penalty issued by a Canadian court was in a case called Reddy v. Saroya, which is a 2026 Alberta Court of Appeal decision. It was $17,550.”
His Courtready website hosts an ever-growing database that tracks AI hallucination cases and sanctions for AI hallucinated cases and misuse. The databases, specifically an article that Zheng wrote about them and the insights they offer, were referenced by the tribunal in its decision.
“There were seven such decisions in 2024, 87 in 2025, and we are already at 74 in the first six months of 2026 alone,” he says. So far, 81 percent of these involve self-represented parties, but that leaves 19 percent – or 32 decisions – where lawyers were sanctioned.
Tom Macintosh Zheng
And while it’s common knowledge that generative AI systems and LLMs “hallucinate” or fabricate facts, lawyers still seem to be including AI-generated content in legal documents without fact-checking. As to why that’s happening, Zheng doesn’t have an answer, despite the collected data.
“I’m a lawyer myself. We were always taught, whenever we do research, we have to verify, and so these are now instances that you see in these 19 percent of cases where the lawyer failed to verify, and they absolutely should have.”
The fact, however, that courts are issuing costs and imposing penalties is evidence to Zheng that “courts are alive to this issue, and they expect more from represented parties, because lawyers are supposed to verify, given that they’re officers of the court, so naturally they have a professional obligation not to mislead the court.” The potential consequences are also increasing: in one matter, a Toronto lawyer now faces criminal contempt proceedings over AI use after admitting to misleading the court.
It’s not a practice that Zheng expects to see go away, even with increased penalties. In fact, he expects the problem to get worse.
He explains that earlier versions of generative AI systems weren’t as proficient at creating plausible-sounding citations. For example, they wouldn’t include neutral citation numbers or years. Or they’d have generic-sounding case names like Smith v. Doe. The most up-to-date AIs, however, can now produce much more realistic-sounding citations. An appendix to the tribunal decision listing Mazaheri’s fabricated citations demonstrates just how realistic they seem. Notes from the LSO didn’t actually say “this is a fabricated citation,” but instead asked the question “Does the cited decision exist?”
Beyond just making up non-existent citations, Zheng says that the AIs also tend to offer real cases and citations, but ones that are irrelevant to the matter at hand. This, too, is seen in the appendix notes, where the LSO had to ask questions such as “What is the relevance of this rule to the proposition asserted in the factum?” or “What is the relevance of this paragraph to the proposition asserted in the factum?” in response to the citations provided by Mazaheri. The same pattern surfaced recently when a judge referred a lawyer to the LSO over a factum that cited real cases but included fabricated quotes, underscoring that the problem is not limited to wholly invented citations.
Still, these are just the baby steps of how AI could be – and is being – misused in preparing legal documents and arguments. Zheng says people are also starting to use them to create fictitious evidence.
In one British Columbia tribunal case, he says, an applicant used AI to describe the signs of a horse needing dental work and to summarize a clinic's veterinary records, then submitted the output as proof. In another case, before the Quebec Labour Tribunal, a worker submitted a large body of AI-generated research to argue that his injury was work-related, which the tribunal dismissed entirely.
Zheng hopes that larger penalties and costs awards will serve as a deterrent, but that won’t put a complete stop to the problem. He knows that courts and law societies have been issuing practice directions and guidance, but he says even those efforts are unlikely to be enough. The LSO and CBA provide guidance to firms to weigh before adopting AI tools, and a recent Canadian Lawyer survey shows how many courts view AI.
“We all know that this issue exists,” he says, “but we haven't really asked the makers of these tools why they continue to hallucinate and to create these fictitious quotes, citations, and cases or propositions of law. I think that’s the next question, and it’s a harder question to ask: Your tools hallucinate. What are you doing to ensure that these tools are not doing what they aren’t supposed to do?”
Additionally, Zheng believes the federal government should play a larger role in regulating AI technology.
He says he was surprised that Canada's national AI strategy, branded "AI for All," does not identify the justice system and legal sector as a key area requiring investment and education, given how real the issue of AI in the legal system has become. He hopes the federal government will devote more attention to it, since the problem is already live, ongoing and set to accelerate.
