An Ontario lawyer has been suspended from practising law for six months after she filed a court document containing fabricated case citations generated by ChatGPT, then compounded the error by repeatedly lying to a judge and to her regulator about what had happened. The July 16, 2026 ruling from the Law Society Tribunal, Law Society of Ontario v Lee, is one of the most detailed disciplinary decisions yet on lawyers’ use of generative AI, and a cautionary tale about how a technology mistake can spiral into a much bigger integrity problem.
Mary Hyun-Sook Lee, a sole practitioner called to the bar in 1995, was representing a client in a family law dispute that had turned into an estate matter after the client’s husband died. Working essentially for free after the client’s Legal Aid funding lapsed, Lee used ChatGPT to help draft a factum, a written legal argument, for an urgent motion.
The factum cited four cases. One was real but was cited backwards: Lee’s factum claimed the court had removed an estate trustee for misconduct, when in fact the court had done the opposite and rejected the misconduct allegations entirely. The other three cases didn’t exist at all. Hyperlinks in the document either led to unrelated rulings or produced error pages.
When the motion was heard in May 2025, the presiding judge couldn’t locate the cases and asked Lee directly whether she’d used ChatGPT. She said no. Asked if a staff member had used it, she said she couldn’t say. Both answers were false, Lee had written the factum herself.
Digging the hole deeper
The judge allowed the motion to proceed but flagged the fabricated citations in his written endorsement and ordered Lee to show cause as to why she shouldn’t be held in contempt of court.
Rather than come clean, Lee escalated the deception. In a letter to the court she claimed that “inquiries with my staff” had revealed a staff member used the AI tool. At a subsequent case conference, she told the judge more specifically that a law student had prepared the factum with ChatGPT without her instruction, and that she’d been “shocked” to learn AI had been used. In reality, Lee had employed neither an associate since 2020 nor a law student since mid-2024. She had done the work, and used the AI tool, entirely on her own.
The judge accepted her account, ultimately ruling that her apparent contrition purged any contempt, and noted, not yet knowing the fuller story, that she had already suffered public shaming as the first lawyer to file AI-hallucinated cases before Ontario’s Superior Court.
Caught by her own regulator
The Law Society of Ontario opened an investigation once the judge’s endorsement became public. Lee initially repeated the same false version of events in writing, again attributing the AI use to an unnamed staff member. When investigators asked for that staff member’s contact details, Lee refused to provide them. Only after being told she was legally obligated to answer did she finally admit, more than a month later, that she alone had drafted the factum and used ChatGPT.
From that point, the Tribunal’s decision notes, Lee cooperated fully, admitted her misconduct, and expressed what regulators accepted as genuine remorse.
A joint proposal the Tribunal almost rejected
The Law Society and Lee’s lawyer jointly proposed a six-month licence suspension plus $10,000 in costs. Under Canadian legal precedent, tribunals are expected to defer to such joint proposals unless the proposed penalty would bring the administration of justice into disrepute.
The panel initially balked. It issued an interim ruling flagging that Lee’s conduct might trigger the “Bolton principle”, the doctrine, drawn from English case law, that the most serious proven dishonesty by a lawyer should result in losing their licence entirely (revocation), not just a suspension. The panel asked both sides for further submissions on whether six months was really proportionate, given that Lee had lied to the court on three separate occasions over roughly two weeks, including twice after she’d had time to reconsider and tell the truth.
In the end, the panel accepted the joint submission, citing a mix of aggravating and mitigating circumstances: the dishonesty was serious and deliberate, but Lee was a sole practitioner nearing the end of an unblemished 30-year career, had taken on the underlying case pro bono for a vulnerable client, had no prior discipline history, and, crucially, had almost no prior experience with AI tools and did not initially know the citations were fabricated when she first argued the motion. She is also facing a separate, ongoing contempt-of-court referral.
Misconduct
The Tribunal was explicit that using AI to help draft legal documents is not, by itself, misconduct. The problem was Lee’s failure to verify what the tool produced before putting her name on it and submitting it to a court, and then her decision to lie about it, repeatedly, rather than admit the mistake early. As the panel put it in adopting the trial judge’s own list of a lawyer’s duties: it is a lawyer’s job to ensure human review of anything generated by AI, and it is “the litigation lawyer’s most fundamental duty not to mislead the court.”
The ruling adds Ontario to a small but growing list of jurisdictions confronting AI-hallucination cases in legal practice, following an earlier, less severe Colorado disciplinary case involving a junior lawyer who also cited fake AI-generated cases. The panel noted the two cases aren’t directly comparable, Lee was a much more senior lawyer, and unlike the Colorado lawyer, she did not disclose the problem voluntarily.
For law firms and solo practitioners increasingly relying on generative AI for research and drafting, the case underscores a now-familiar warning from courts and regulators worldwide: the tool’s output is only a starting point, and the professional obligation to check it, and to be honest when things go wrong, remains entirely human.
