This article is general information, not legal advice, and reflects Ontario information as of October 19, 2026. It is based on press and law-firm reports of the Tribunal’s order and reasons (linked below); we have not read the reasons ourselves, so confirm any detail against the decision, cited in those reports as 2026 ONLSTH 136. We do not name the lawyer, who is described in the reports as a Toronto sole practitioner.
What happened
According to the Law Times report on the reasons (July 21, 2026), the lawyer had about 30 years at the Bar and no discipline history, and was acting in a family law matter. A Gardiner Roberts LLP summary (July 24, 2026) says the factum linked four cases: three did not exist, and the fourth was described as having an outcome opposite to the one the factum stated.
The reports say the lawyer first told the motions judge she had not used ChatGPT, then said staff had, and later admitted she had prepared the factum herself. The Law Times report says she repeated the staff explanation to the Law Society. The judge referred a contempt proceeding to the Attorney General, which the report says is ongoing, separate from the Tribunal matter.
The six-month suspension, which Law Times reported on July 16 as starting July 17, was a joint submission from the lawyer and the Law Society that the Tribunal accepted. Law Times also reports counsel told the Tribunal they could find no Ontario penalty decisions involving a lawyer or paralegal relying on AI-generated material containing hallucinations.
What the Tribunal said
| Point | As reported |
|---|---|
| Is AI use misconduct? | Not, per se. Law Times reports the Tribunal said the error was failing to verify the accuracy of the factum. |
| The honesty findings | The Tribunal was “extremely troubled” by repeated lies to the court and the Law Society, which it called serious integrity failures (Law Times). |
| Rules cited | Per Gardiner Roberts: Rule 3.1-2 (competence) for the non-existent and unsupportive authorities, Rule 5.1-2(e) for knowingly deceiving the court, Rule 2.1-1 for misleading the Law Society. |
| Mitigation | Remorse, a long record with no discipline, almost no prior AI experience, and a presiding judge who accepted her apology (Law Times). |
| Cost | Six months out of practice and $10,000 payable to the Law Society, on a joint submission. |
Neither report records the Tribunal giving firm-policy recommendations. What follows is our reading of what the facts imply, not the Tribunal’s advice. Read it alongside the Law Society’s own generative AI guidance.
Stay ahead of the next compliance change.
One email when a rule, a deadline, or a product update actually affects how you work — nothing else.
Four controls the facts point to
- A named human verifies every authority before filing. Not “someone checked”: a person opens each case, confirms it exists, and confirms it says what the factum says. The reported failure included a real case described with the opposite outcome, so existence checks alone are not enough. Gardiner Roberts notes Rule 4.06.1 of the Rules of Civil Procedure requires counsel to certify the authenticity of each authority cited. Our citation verification checklist turns this into a sign-off step.
- A written rule for what to say when a judge or the Law Society asks. The integrity findings came from the answers, not the tool. The policy should say: if asked whether AI was used, answer accurately and immediately, and correct the record the moment you learn of an error. See the disclosure duty for AI in Ontario court filings and the courts’ disclosure rules tracker.
- A record of who drafted what. The reports describe shifting accounts of who prepared the factum. A simple drafting log (author, tools used, reviewer) makes the accurate answer the easy one. If your firm supervises articling students or junior staff, the supervision checklist covers who may use which tool.
- Training before access. The reports say the lawyer had almost no prior AI experience and did not know hallucinations existed until the judge raised them. A short briefing on how these tools fail, given before anyone is allowed to use them on a live file, costs far less than the alternative. Our tracker of fabricated citations in Canadian courts and the Law Society’s earlier costs order give you real examples to use.
What this does not tell you
This was a joint submission, so the sanction reflects a negotiated outcome rather than a contested hearing, and Law Times notes the lawyer’s dishonesty may have influenced it. Do not read six months as a tariff for hallucinated citations alone. The Tribunal’s own reasons are the source to check before relying on any of this in a policy or a submission.