AI Regulation Tracker / Enforcement action
Ontario Law Society Tribunal Orders Full $31,150 Costs Over Hallucinated AI Citations
In a decision released June 12, 2026 and surfaced publicly around July 9 through Law Times, the Law Society Tribunal ordered a self-represented licensee to pay the full $31,150 in costs, driven by what the panel called his irresponsible use of AI. This is a costs order, not a misconduct finding. A separate conduct application is still pending.
An Ontario tribunal has put a specific number on what fabricated AI citations can cost. In Mazaheri v Law Society of Ontario, released June 12, 2026, the Law Society Tribunal ordered a self-represented licensee to pay the full $31,150 in costs after he included hallucinated authorities in his written arguments. The decision drew little attention at first and then spread around July 9, 2026 after Law Times reported it. I am flagging both dates on purpose, because this is a June decision that became news in July, not a July ruling.
What actually happened
The licensee, who represented himself, brought two motions before the Tribunal. One asked the Tribunal to cancel an interlocutory suspension that was already in place. The other asked it to exclude evidence filed by the Law Society of Ontario and to have the panel recuse itself. The three-member panel, chaired by David Aterman and sitting with Jason Poliacik and Julian Richards, dismissed both motions.
Then came costs. The Law Society asked for its full costs of $31,150, and the Tribunal awarded the entire amount, payable immediately, with no reduction at all. What pushed the panel to a full award was the licensee's conduct in how he argued the motions. He had put fabricated case citations, the kind a generative AI model invents when it does not have a real authority to cite, into his materials.
The panel was blunt about it. At paragraph 8 it wrote that "his irresponsible use of artificial intelligence not only wasted time, cost, and effort, it was profoundly improper. The respondent should bear the full cost of his actions. The rest of the professions should not pay for this kind of behaviour." That last line is the reasoning behind the number. The Tribunal did not want the cost of cleaning up fabricated citations to fall on everyone else who practises properly.
A first for this tribunal, and a number worth reading closely
The panel noted that this was new ground for it. At paragraph 35 it said that "as far as we know, this is the first time that a party has included 'hallucinated' authorities in arguments before the Law Society Tribunal." So the $31,150 is not the Tribunal following a settled tariff for AI misuse. It is the Tribunal deciding, the first time it saw this, that a full-cost award was the right response.
The decision also set the case in a wider context, and this is the part US readers should sit with. The panel cited CanLII data on fabricated-citation cases in Canadian legal proceedings: at paragraph 36 it reported "seven in 2024, 86 in 2025, and 39 in the first quarter of 2026, totalling 132." That is a steep climb in a short span. Even more telling for anyone who assumes this is only a self-represented-litigant problem, paragraph 37 records that "in 24 of 132 cases, a represented party submitted hallucinated citations." Roughly one in five of these fabricated-citation incidents came in through a party that had a lawyer.
What this is, and what it is not
I want to be careful here, because the headline number invites over-reading.
This is a costs order. It flows from two motions the licensee brought and lost, and the Tribunal used its costs discretion to make him pay the full amount because of how he conducted them. It is not a finding that he committed professional misconduct. The panel did not rule on the underlying conduct application against him, and that application is still pending. Do not read the $31,150 as a discipline penalty. It is the cost of the motions, awarded in full.
It is also a Canadian tribunal decision. Inside Ontario it binds these parties and carries weight before the Tribunal going forward. Outside Canada it is persuasive at most. A US judge weighing a fabricated-citation problem is not required to follow anything the Law Society Tribunal said, and could cite it only as an example of how another serious adjudicator handled the same behaviour.
And the timing matters for how you frame it. The decision is dated June 12, 2026. It became a talking point in the second week of July when Law Times covered it. If you cite this, cite it as a June decision, not a July one.
What this means for US lawyers
The reason this cross-border decision belongs on a US lawyer's radar is not that it binds you. It does not. The reason is the pattern it captures.
US courts have been sanctioning lawyers over fabricated AI citations for two years now, and the counts keep climbing. This Ontario decision adds a clean, quotable data point from a serious adjudicator, with a specific dollar figure and a stated rationale: the profession should not absorb the cost of one person's irresponsible AI use. That reasoning travels. It is close to how US judges have framed their own sanctions.
The sharper lesson is in paragraph 37. The instinct is to treat hallucinated citations as a hazard of self-represented parties who do not know better. The CanLII data cuts against that. Nearly one in five of these incidents came through a represented party, which means a lawyer's name was on filings that contained authorities no one had opened. That is the failure mode to guard against in your own shop: an associate or a contract lawyer runs a research task through a model, the model invents support, and it goes out under a signature without anyone pulling the cases. The fix is not complicated, but it has to be a rule, not a hope. Every citation gets verified against the actual source before it is filed.
What to do now
Read the decision for what it is: a full costs award over fabricated citations, not a misconduct ruling, from a Canadian tribunal that is persuasive at most in the US. Treat the CanLII trend line, 7 in 2024, 86 in 2025, and 39 in the first quarter of 2026, as the real signal, because it shows the volume of these incidents rising fast. Take the represented-party finding seriously and build a verification step into your workflow so that no cited authority reaches a filing without a human confirming it exists and says what the brief claims. Keep a named person accountable for citation checks on anything AI touched. And if you reference this case in your own work, get the citation and the date right: Mazaheri v Law Society of Ontario, 2026 ONLSTH 112, decided June 12, 2026.
Questions professionals are asking
Was this a misconduct finding against the licensee?
No. This is a costs order on two interlocutory motions the licensee brought and lost. The Tribunal awarded the Law Society its full $31,150 in costs and cited his AI use as the reason for a full award. It did not rule that he committed professional misconduct, and a separate conduct application against him is still pending.
When was the decision actually made?
It was released June 12, 2026. It became widely visible around July 9, 2026 after Law Times reported it. It is a June decision that became news in July, not a July ruling.
Does this bind US courts or US lawyers?
No. It is a decision of the Law Society Tribunal in Ontario, Canada. It binds the parties in that matter and carries weight before the Tribunal. In a US court it is persuasive at most, useful as an example rather than as authority.
What is the represented-party point everyone is citing?
The Tribunal cited CanLII data showing 132 Canadian cases involving fabricated citations, counted as seven in 2024, 86 in 2025, and 39 in the first quarter of 2026. In 24 of those 132 cases, a represented party, meaning a party with a lawyer, submitted the hallucinated citations. That is roughly one in five, which shows the problem is not limited to self-represented litigants.
What should a US firm take from this?
Treat it as a data point in the broader fabricated-citation sanctions wave, and act on the represented-party finding. Build a verification step into your workflow so no cited authority reaches a filing without a person confirming it exists and supports the point. Keep a named person accountable for checking any citation that AI touched.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any decision or rule applies to your situation with qualified professionals in the relevant jurisdiction.