What the tribunal actually punished
On July 16, 2026, the Law Society Tribunal suspended Toronto family law practitioner Mary Hyun-Sook Lee, known professionally as Jisuh Lee, for six months and ordered her to pay C$10,000 in costs. The case, Law Society of Ontario v. Lee, 2026 ONLSTH 136, is being described as Canada's first disciplinary decision to sanction a lawyer specifically for misusing generative AI in court filings. Lee had used ChatGPT to prepare a factum containing hyperlinks to four cases; three did not exist, and the one real case was mischaracterized as having reached the opposite conclusion from what it actually held. When the motions judge, Justice Frederick Myers, asked her directly whether she had used AI to write the document, Lee said no. She then told the court a law student had prepared it without her knowledge — a claim that was false, since she had employed no staff since 2020. Only once the Law Society opened an investigation did she admit she had written the factum herself, with ChatGPT's help, under time pressure.
The distinction the tribunal drew matters more than the headline. As reporting on the decision makes clear, the panel did not hold that using AI to help draft court documents is itself misconduct — it held that Lee failed to verify what the tool produced, and then lied about it three separate times: to the judge, in a follow-up letter, and to Law Society investigators. That sequencing is the whole story. A lawyer who cites a hallucinated case by mistake has a competence problem. A lawyer who denies using AI after being asked directly, under oath-adjacent circumstances, has an honesty problem — and honesty is the one thing law societies treat as close to a strict-liability offense.
The steelman for tougher AI rules
The case for a firmer regulatory hand here is not hard to make. Citing fabricated case law wastes judicial time, forces opposing counsel to chase down authorities that don't exist, and — where it goes undetected — can produce rulings built on law that was never actually decided. Ontario's Superior Court of Justice evidently agreed the problem was serious enough to act on structurally: on February 12, 2026, it issued new practice directions on the responsible use of AI covering civil, family, and criminal proceedings, aimed at what the court called promoting "transparency, accuracy, and accountability." The Federal Court moved even earlier, issuing interim AI guidelines on September 29, 2025 that require litigants to disclose when generative tools were used in preparing their materials. Those aren't overreactions. A profession whose entire product is trustworthy assertions about what the law says has a legitimate interest in making sure a chatbot's confident fabrications don't quietly become the record.
Why blanket AI bans would be the wrong response anyway
But Lee's case is also useful evidence against the harder-edged proposals now circulating — outright bans on AI-assisted drafting, mandatory AI-use disclosure enforced with sanctions regardless of accuracy, or treating any hallucinated citation as presumptive misconduct. Lee's own conduct shows the existing framework already worked: the Rules of Professional Conduct she was found to have breached — Rule 3.1-2 (competence), Rule 5.1-2(e) (candour to the tribunal), and Rule 2.1-1 (integrity) — are decades old and were never written with ChatGPT in mind, yet they captured the misconduct precisely. No new AI-specific offense was needed; verification and honesty were already mandatory. That is the more defensible baseline than category bans: hold lawyers to the duties they already owe, and let the tool remain neutral.
The scale point cuts the same way. Legal-tech researcher Damien Charlotin's tracker, cited in Canadian Lawyer's August 2026 review of the case law, has now logged at least 150 Canadian court and tribunal decisions addressing AI-hallucinated citations since early 2024 — and the same review found that the resulting body of rulings mostly "offer[s] few new insights," reminding lawyers of duties that already existed rather than creating new ones. Courts have consistently distinguished, as British Columbia's Justice Masuhara did in Zhang v. Chen, 2024 BCSC 285 — the case widely cited as Canada's first AI-hallucination episode — between a lawyer who makes a good-faith error and discloses it promptly, and one who conceals it. In Zhang, counsel withdrew the fabricated citations, testified she hadn't known ChatGPT could invent case law, and avoided a special costs order, though she still paid ordinary costs for the wasted time. Lee's case sits at the opposite end only because she chose concealment over correction, not because AI use itself crossed a line the earlier case hadn't already marked.
The policy takeaway
The sensible regulatory response, visible in Ontario's own February practice directions and the Federal Court's disclosure rule, is procedural: require lawyers to know what tools they used and be able to answer honestly when asked, verify every citation before it reaches a judge, and reserve serious sanction for concealment and dishonesty rather than for the underlying act of drafting assistance. That calibrates punishment to the actual harm — a false statement to a court — rather than to the presence of a technology that, used carefully, is a legitimate research aid. Ontario's tribunal got the balance right: it suspended a lawyer for lying, not for typing a prompt.