Ko v. Li, 2025 ONSC 2965; 2025 ONSC 6785
·Ontario Superior Court of Justice·CARegulatory referral
In Ontario, lawyer Jisuh Lee filed a ChatGPT-fabricated factum, then admitted she had lied about authorship after initially blaming others. Justice Myers referred a criminal-contempt prosecution to Ontario's Attorney General, and the Law Society Tribunal later suspended Lee for six months and ordered CAD 10,000 costs for misconduct including dishonesty.
- Court
- Ontario Superior Court of Justice
- Date
- 2025-12-04
- Who used the AI
- Not recorded
- AI tool
- ChatGPT
- Conduct
- Applicant's counsel Jisuh Lee filed a factum citing fabricated cases hallucinated by ChatGPT and relied on them orally. An initial contempt was treated as purged after she blamed staff (2025 ONSC 2965, 20 May 2025); Lee then admitted, in a 30 Sept 2025 letter, that she had drafted the factum herself and had lied about it out of fear and embarrassment — a fresh deception of the court (2025 ONSC 6785, 4 Dec 2025).
- Outcome
- Regulatory referral
- Penalty
- No monetary penalty recorded
- Regulatory outcome
- Law Society of Ontario: The Law Society Tribunal found professional misconduct, including incompetent AI use and dishonesty to the court and the Law Society, and suspended Mary Hyun-Sook Lee for six months with CAD 10,000 costs. The separate criminal-contempt process before the Ontario Superior Court remains unresolved on the sources checked. (2026-07-16)
- Status
- Verified against listed sources
- Last checked
- 2026-09-04
What was fabricated or misused
- Johnson v Lanka, 2010 ONSC 4124 — cited for removing an estate trustee; the case stated the opposite
- Meschino Estate v Meschino, 1998 Can LII 14734 (ONSC) — non-existent case
- Alam v Shah, 2023 ONSC 1772 — non-existent case
- DaCosta v DaCosta, 2010 ONSC 2178 — non-existent case
What the court said
“To support her legal submissions, Ms. Lee included in her factum references to several non-existent or fake precedent court cases. In her oral argument in open court before me on May 1, 2025, Ms. Lee relied on two non-existent precedent cases from her factum.”
“She then deliberately misled a court multiple times when asked about her use of AI.”
“The failure to perform legal services competently in this case lies in the Lawyer’s failure to verify the authenticity of the authorities generated by AI.”
“The parties jointly proposed as penalty a suspension of six months. We reserved on the penalty and costs. After considering the evidence and submissions, including post-hearing written submissions, we accept the joint proposal and order a suspension of six months.”
Timeline
- Urgent motion was originally scheduled; later adjourned to 1 May 2025.
- Lee served a motion factum with four AI-hallucinated case authorities.
- At the motion, Myers J questioned the cases and asked whether ChatGPT was used.
- Myers J ordered Lee to show cause why she should not be cited for contempt.
- Show-cause proceeding dismissed on education and no-billing conditions after Lee’s explanations.
- Lee wrote to Myers J apologising for misrepresentations about who used ChatGPT.
- Law Society Tribunal ordered a six-month suspension and CAD 10,000 costs.
Why this case matters
Ko v. Li separates the hallucination from the later candour breach. Zhang v. Chen had already shown that a Canadian lawyer could face personal costs and file review after citing ChatGPT cases. Ko escalated because Lee first denied using ChatGPT, then blamed staff, and later admitted she had written the factum herself. The Law Society Tribunal treated the AI misuse and the dishonesty as separate professional misconduct and imposed a six-month suspension. That makes the record less about the tool than the lawyer’s response after the court asked direct questions.
Practice note
Ontario counsel must comply with Rule 4.06.1 by certifying that each authority in a factum is authentic and by giving pinpoint references. The check has to be personal or supervised, not assumed from ChatGPT output. If a false citation reaches court, answer questions candidly, correct the record at once, and tell the regulator the same account.
Primary sources for this incident
Related regulation records for this incident
Read our full analysis of this case →
Similar incidents on the record
- Zhang v Chen, 2024 BCSC 2852024-02-20 · Supreme Court of British Columbia · Costs orderThe first reported Canadian case of AI-hallucinated citations: a Vancouver family lawyer cited two ChatGPT-invented cases in a notice of application, was ordered to pay costs personally, had her other files reviewed, an…
- Mezu v. Mezu (Md. App. Ct. 2025), No. 361 Sept. Term 20252025-10-29 · Appellate Court of Maryland · Regulatory referralMaryland's intermediate appellate court referred an attorney to the state Attorney Grievance Commission after his client's brief cited multiple fictitious ChatGPT-generated cases and he conceded he routinely did not rea…
- MS (Professional conduct, AI generated documents) Bangladesh, [2025] UKUT 305 (IAC)2025-08-12 · Upper Tribunal (Immigration and Asylum Chamber) · Regulatory referralBefore the Upper Tribunal (Immigration and Asylum Chamber), barrister Muhammad Mujeebur Rahman cited a fabricated Court of Appeal case, Y (China), drafted using ChatGPT without verification. The tribunal, led by its Pre…
- Valu v Minister for Immigration and Multicultural Affairs (No 2) [2025] FedCFamC2G 952025-01-31 · Federal Circuit and Family Court of Australia (Division 2) · Regulatory referralIn one of Australia's first AI-citation cases, a solicitor before the Federal Circuit and Family Court cited non-existent Federal Court authorities and fabricated tribunal quotes generated by ChatGPT. Judge Skaros found…
Cite this record
SafeLegalAI Global Legal AI Incident Tracker, “Ko v. Li, 2025 ONSC 2965; 2025 ONSC 6785”, safelegalai.com/tracker/ko-v-li (accessed 2026-09-16). Data: CC BY 4.0.