← All case files
verified deployment legal · CA · cross

Ontario Law Society Tribunal orders a lawyer to pay $31,150 in costs after Grok-generated fake citations

In Mazaheri v Law Society of Ontario (2026 ONLSTH 112), the Law Society Tribunal ordered a suspended lawyer to pay the regulator's full costs of $31,150 after he filed motion materials produced by generative AI (which he admitted was 'in particular Grok') that cited non-existent decisions and misused real ones. The panel called the AI misuse 'an additional and significantly aggravating factor' and noted it was the first time a party had put hallucinated authorities before the Tribunal.

MetricBeforeAfter
$31,150 (CAD) in costs ordered against the respondent, payable by June 26, 2026, with 4% interest on any overdue amount
Full costs awarded without any reduction; AI misuse held to be 'an additional and significantly aggravating factor'
First time a party has included 'hallucinated' authorities in arguments before the Law Society Tribunal

The problem

Shahryar Mazaheri, an Ontario lawyer whose licence had been suspended on an interlocutory basis on November 12, 2024, brought motions before the Law Society Tribunal to cancel or vary that suspension and, later, to exclude the Law Society’s evidence and have the panel recuse itself (source). The factum, supplementary factum, affidavit and supplementary affidavit he filed, dated November 14, 2025, were in the panel’s words “produced with the assistance of generative artificial intelligence” and full of “references to non-existent decisions, or decisions that exist but do not stand for the propositions outlined by him” (source). The same materials pressed the Tribunal’s Rules of Practice and Procedure into service as substantive law, and the panel called the argument “completely incoherent” (source).

What was built

Nothing was deployed: this is an AI-governance record, not a vendor win. After the panel found the materials “appeared to contain numerous hallucinated citations and wrong or misleading propositions of law,” it issued a case-management direction on November 25, 2025, convened a case-management hearing, and permitted the respondent to re-file (source). Mazaheri wrote to the Tribunal on November 30, 2025 admitting he had used generative AI, stating the errors “arose for [sic] my over-reliance on generative artificial intelligence tools in particular Grok to assist me in researching and drafting the documents while I was trying to manage filing these materials on my own” (source). The panel made no finding as to which product caused the fabrications; Grok is named only in the respondent’s own admission (source).

The outcome

Both motions failed (source). On June 12, 2026 the three-member panel, Paul Aterman as chair with Lubomir Poliacik and Michelle Richards, ruled on the costs of the motions (source).

The number that gets quoted: $31,150 in costs, awarded in full. The panel ordered that “the respondent will pay costs to the Law Society in the amount of $31,150 by June 26, 2026,” with interest accruing “on any overdue part of these costs at a rate of 4% per year,” and awarded the Law Society’s full costs without any reduction (source). It held that his “irresponsible use of artificial intelligence not only wasted time, cost, and effort, it was profoundly improper” and was “an additional and significantly aggravating factor when we consider his conduct in these motions” (source). Two independent Canadian newsrooms restated the figure the same week: Investment Executive reported the panel “awarded the Law Society the full costs that it sought on those motions, $31,150” (source), and Law Times reported the Tribunal “ordering costs of $31,150 to the Law Society of Ontario” (source).

The context the panel set: a first for this Tribunal, inside a rising trend. As far as the panel knew, this was the first time a party had put hallucinated authorities in arguments before the Law Society Tribunal, though “by no means a first in Canadian law” (source). It noted that CanLII-indexed cases where a court or tribunal identified that a party had submitted a fictitious citation rose from seven in 2024 to 86 in 2025 and 39 in the first quarter of 2026, “totalling 132” (source). The decision expressly states the panel “make[s] no determination as to whether the respondent has engaged in professional misconduct,” which a separate panel will decide (source).

How this was verified

This case carries its badge under TIN’s independent-origin standard: the figures come from a published Law Society Tribunal decision (2026 ONLSTH 112), an independent adjudicator rather than a self-reporting party, read in full on 2026-08-03 and byte-tied to the archived copy of the decision (source). No client-confirmation call is needed to authenticate a published tribunal decision, and none was made. The $31,150 costs award, the 4% interest, the June 26, 2026 deadline and the CanLII trend figures were each read back to the decision text, and two independent Canadian newsrooms corroborate the headline figure. The one limit worth stating plainly: the panel made no finding as to which product caused the fabrications, so the attribution to Grok rests solely on the respondent’s own written admission, not on any technical determination by the Tribunal.

The same Commonwealth pattern, an AI-fabricated-citation filing answered with a full costs order against the person who filed it, appears in the Federal Court of Australia’s indemnity-costs order in Murray (Wamba Wemba) v Victoria. For the European counterpart, Italy’s Court of Cassation fined an appellant 5,000 euro for AI-hallucinated citations, another apex-level body treating the trend as a discipline-and-costs question. And in the United States, a federal court fined a lawyer $6,000 for three briefs full of AI-fabricated citations, the same escalation from warning to sanction in a different legal system.


Sources

Legend: Tier 1 = primary (the tribunal’s own published decision); Tier 2 = independent press naming the parties. Every figure was checked against the live decision on 2026-08-03.

  1. Law Society Tribunal (Hearing Division), “Mazaheri v Law Society of Ontario, 2026 ONLSTH 112,” 2026-06-12 (Tier 1, the published decision on costs). https://www.canlii.org/en/on/onlst/doc/2026/2026onlsth112/2026onlsth112.html
  2. Law Society Tribunal, decision PDF, mirror of 2026 ONLSTH 112 read in full for this check, 2026-06-12 (Tier 1, primary document copy). https://websitedc.s3.amazonaws.com/documents/Mazaheri-v-Law-Society-of-Ontario-2026-ONLSTH-112.pdf
  3. Investment Executive (Newcom Media), “Faulty AI used in legal tribunal,” 2026-06-15 (Tier 2, independent newsroom). https://www.investmentexecutive.com/news/regulation/faulty-ai-used-in-legal-tribunal/
  4. Law Times (Key Media), “Lawyer who used AI-fabricated citations hit with $31,150 in costs to LSO,” 2026-06-17 (Tier 2, independent newsroom). https://www.lawtimesnews.com/resources/professional-regulation/lawyer-who-used-ai-fabricated-citations-hit-with-31150-in-costs-to-lso/393407

Generative AI (Grok, per the respondent's admission)

Verification record
Status
verified
Method
Independent-origin tribunal decision (2026 ONLSTH 112), byte-tied to a Wayback capture; corroborated by two independent Canadian newsroom publishers (Newcom Media, Key Media).
Verified on
2026-07-31
Provider
Generative AI (respondent's admission names Grok / xAI; the decision refers generically to 'generative artificial intelligence')
Client
Law Society Tribunal (Hearing Division), Ontario: Mazaheri v Law Society of Ontario, 2026 ONLSTH 112 · Legal profession regulation / courts
Disclosure
named