analysisAI HallucinationsAI in Courts

Ko v. Li: how a ChatGPT factum became a criminal-contempt referral in Ontario

An Ontario lawyer filed a ChatGPT-fabricated factum, then lied about writing it — and drew a criminal-contempt referral to the Attorney General.

OpenAI ChatGPT

Edited and verified by Cognesio LLP

Researched with AI assistance · sources verified by Cognesio LLP · How this was made ↓

Correction (2026-09-04): The Law Society Tribunal has since issued a disposition in Law Society of Ontario v Lee, 2026 ONLSTH 136, suspending Lee for six months and ordering CAD 10,000 costs. The criminal-contempt process remains unresolved on the sources checked.

In Ontario, a lawyer who filed a factum built on cases invented by ChatGPT ended up facing a criminal-contempt prosecution — not for the fabricated citations, but for lying about them afterwards. Ko v. Li, decided in the Ontario Superior Court of Justice by Justice Myers across a series of endorsements (principally 2025 ONSC 2965 and 2025 ONSC 6785), is the tracker’s clearest example of the cover-up carrying a heavier consequence than the original error.

What happened

Applicant’s counsel Jisuh Lee filed a factum that cited non-existent cases and relied on them in oral argument at a May 2025 motion. The problem surfaced when Justice Myers could not find the cases online. Asked directly whether she had used ChatGPT to write the factum, Lee answered, “I did not.” The court ordered her to show cause why she should not be held in contempt.

At a case conference days later, Lee said staff were responsible — that she had delegated the factum to a student and learned only afterwards that AI had been used. On that basis the court treated the initial contempt as purged, satisfied that her expressions of accountability and regret, together with withdrawal of the offending factum, had cleared it. It was the off-ramp other AI-citation cases have offered a lawyer who owns the mistake.

The matter did not end there. After the Law Society opened an investigation, Lee wrote to the court on 30 September 2025 to correct the record. She had, she admitted, drafted the factum “solely and entirely by myself,” had used ChatGPT, had not verified the citations, and had blamed staff “out of fear of the potential consequences and sheer embarrassment.” The deception, rather than the hallucination, was now the problem.

What the court decided

Justice Myers treated the admission as a fresh wrong rather than a reason to reopen the first proceeding. As he put it in the October 2025 letter that began the new process:

“Your admission of dishonesty in the contempt proceeding is a fresh act. The initial contempt proceeding concerned principally your failure to review the citations in your factum before delivering it and relying on fake cases in open court. The new disclosure raises issues of deception of the court in a contempt of court proceeding.”

He ordered Lee to show cause why she should not be held in criminal contempt for obstructing the administration of justice, and referred carriage of the prosecution to the Attorney General of Ontario. By the December 2025 case conference the Crown had taken carriage, and the court appointed an amicus curiae to help ensure a fair process. No finding of contempt has been made. Myers J set out how far outside the ordinary run the case sits:

“I have not seen any case law in which a lawyer, owing duties of candour and honour, admits to deliberately misleading a court in a criminal contempt of court proceeding about herself. This is a very unusual case and one that may resonate throughout the administration of justice.”

The regulatory track

Running alongside the contempt process, the Law Society of Ontario investigated Lee’s conduct (file CAS-165781-ZOC5S2) and attended the December case conference as an observer. On 16 July 2026, the Law Society Tribunal found professional misconduct and ordered a six-month suspension plus CAD 10,000 costs. The criminal-contempt process remains unresolved on the sources checked.

Why it matters

Ko v. Li separates two failures that AI-citation cases usually run together. The first, filing unchecked fabricated cases, is by now familiar and often draws costs or a referral. The second, misleading the court about it, is what drove this case toward criminal contempt.

The pattern echoes the Oregon sanctions in Couvrette v. Wisnovsky, where concealment rather than the fabrication itself pushed the court to its most severe response. For a lawyer, the practical lesson sits entirely in the second act: a hallucinated citation is a serious error, but the response to being caught is what determines how far the case escalates.

Sources