Disclosure: DROZlegal publishes this article and builds a practice-automation product for Canadian law firms. Facts about Mazaheri v. Law Society of Ontario and the Law Society of Ontario's 2026 budget are drawn from Law Times News reporting (both articles independently re-verified against their published text this week), cross-checked against Mondaq, Canadian HR Reporter, and CanLII's own case listing, each cited by name and date below — verify current details directly against those sources before relying on them.
A record costs order, and what actually happened
Shahryar Mazaheri's law licence had already been suspended since November 2024 when he filed two motions before the Law Society Tribunal, self-represented: one to cancel or vary that interlocutory suspension, and a second to exclude evidence the Law Society intended to rely on. The submissions supporting both motions — prepared with the help of a generative AI tool, Grok, whose output he did not check — contained citations to cases that, per the Tribunal's own findings, either did not exist or had nothing to do with the points he was making.
The Tribunal did not treat that as a side issue. It characterized the reliance on unverified AI output as an "irresponsible use of artificial intelligence" and treated it as "an additional and significantly aggravating factor" in the costs it awarded — ordering Mazaheri to pay $31,150 to the Law Society of Ontario (Mazaheri v. Law Society of Ontario, 2026 ONLSTH 112, Law Society Tribunal, decision reported June 2026).
The Tribunal found Mazaheri's submissions relied on "cases that don't exist and real cases that had nothing to do with the points" he was making, calling it "an additional and significantly aggravating factor" in the $31,150 costs order. — Mazaheri v. Law Society of Ontario, 2026 ONLSTH 112, reported by Law Times News, June 17, 2026
Trade coverage put the figure in context. Toronto lawyer Tom Macintosh Zheng, co-founder of Courtready.ca, told Law Times that until this case, the largest AI-citation penalty issued by a Canadian court was $17,550, from Reddy v. Saroya, a 2026 Alberta Court of Appeal decision — putting the Mazaheri costs order nearly double the prior high, and, in his telling, the largest such award to date.
Two precisions worth keeping straight. This is a costs order, not a fine and not a court finding of contempt — and the Law Society Tribunal adjudicates independently of the Law Society's regulatory arm, with the LSO appearing here as the opposing party and the costs ordered payable to it. Nor is this the first Canadian incident of an AI-fabricated citation reaching a legal proceeding: we've covered an earlier, distinct Ontario Superior Court case, Ko v. Li, in our guide to AI legal research software, and won't repeat those facts here. What's new is the dollar figure, the venue being the regulator's own Tribunal, and the timing relative to what that same regulator was funding.
The same regulator is now funding the fix
Months before the Mazaheri decision, the Law Society of Ontario's board approved its 2026 budget, reported by Law Times News on December 1, 2025. The headline number is a modest licensee fee increase: annual fees for practising lawyers rise from $2,039 to $2,080, a 2% increase, and paralegal fees rise from $1,023 to $1,037, roughly 1.4%.
Buried in that same budget is the more interesting line: the 2026 budget goes toward funding CanLII's push to roll out an AI-powered legal research tool — described in the coverage only as "providing access to information." CanLII has not published the tool's architecture, so treat the obvious expectation — that a research tool from the operator of Canada's primary free case-law and legislation database answers from that corpus rather than an open-ended model's memory — as an expectation, not a specification.
The LSO framed the fee increase itself as sitting below the then-current 2.2% inflation rate, with proceeds going to initiatives that add value to the profession, support public protection, and respond to "calls for innovation."
| Development | What happened | Source & date |
|---|---|---|
| Penalize | $31,150 costs order against a suspended lawyer over AI-fabricated citations in Tribunal filings — reported at the time as the largest AI-citation costs award by a Canadian court or tribunal | Law Times News, June 17, 2026 (Mazaheri v. LSO, 2026 ONLSTH 112) |
| Fund | 2026 budget raises licensee fees ~2% (lawyers, $2,039→$2,080) / ~1.4% (paralegals, $1,023→$1,037), with proceeds backing CanLII's generative-AI legal research project among other initiatives | Law Times News, December 1, 2025 |
Neither story references the other in the coverage we reviewed. Read together, they describe one regulator running both sides of the same problem in the same budget year: penalize a lawyer for citations nobody checked, and pay to build a tool designed to make checking easier in the first place.
Why this matters if you're evaluating AI for your practice
None of this changes the underlying duty. The Law Society of Ontario's generative-AI guidance already maps AI use onto three existing professional-conduct rules — competence, confidentiality, and supervision — and the competence rule has always required independently verifying anything a tool produces before it reaches a filing or a client. We've walked through that guidance in full, rule by rule, in our explainer on the LSO's generative AI guidance; Mazaheri is what happens when that verification step is skipped, not a new rule being applied.
What the CanLII funding line adds is the infrastructure side of the same problem. A tool that only answers from a maintained, retrievable case-law and legislation corpus — rather than an open-ended model's recall — is a meaningfully different starting point than one that can invent a plausible-sounding citation with confidence. We've covered what that verified-citation architecture actually requires, and the questions worth putting to any vendor before a citation from their tool reaches a filing, in our guide to AI legal research software for Canadian firms — worth reading before, not after, your firm signs anything.
Three things this pairing confirms, regardless of which AI tool your firm uses:
- The regulator isn't waiting for more incidents to act. A $31,150 costs order in a Tribunal proceeding — nearly double the previous reported Canadian high — on top of an existing suspension, signals this isn't being treated as a minor procedural lapse.
- "Grounded in case law" is becoming table stakes, not a differentiator. When the regulator itself is putting licensee money behind CanLII building a legal research tool, a vendor's open-ended chat model without a named, checkable database is a harder sell than it was a year ago.
- Verification is still the lawyer's job, not the tool's. A better-built research tool reduces how often a fabricated case reaches a lawyer; it doesn't remove the obligation to open the case and read it before it goes in front of a judge or a Tribunal panel.
Update, September 2, 2026. Ontario has since moved from professional-conduct guidance to a procedural rule on exactly this problem. Ontario Regulation 275/26, filed August 12, 2026 and in force since September 1, 2026, amends the Rules of Civil Procedure so that any person signing a factum (Rule 4.06.1) or expert report (Rule 53.03, via a revised Form 53 Acknowledgment of Expert's Duty) must certify they are satisfied any quotation purported to be taken from an authority accurately reflects that authority's text. Source: practicePRO, “Effective September 1, 2026: Amendments to Rules of Civil Procedure (O. Reg. 275/26),” August 2026; filing confirmed in the Ontario Gazette, Vol. 159, Issue 35 (August 29, 2026).
Update, September 3, 2026. The Law Society Tribunal has since gone well past a costs order. On July 16, 2026, it suspended Toronto lawyer Mary Hyun-Sook Lee's licence for six months, effective July 17, and ordered her to pay $10,000 in costs to the Law Society of Ontario — after finding she relied on a factum containing AI-hallucinated case citations from ChatGPT in what began as a family-law matter and became a complex estate matter, then lied to the Ontario Superior Court and to the Law Society about her AI use before admitting, once the Law Society opened its own investigation, that she had prepared the document herself and had employed no staff to blame since 2020 (Law Society Tribunal, 2026 ONLSTH 136, decision dated July 16, 2026). The Tribunal described her integrity lapses as “out of character” against a clean 30-year disciplinary record — language that mitigated the outcome without preventing the suspension. The distinction from Mazaheri is the sanction, not the underlying error: Mazaheri paid $31,150 in costs for AI citations nobody checked; Lee lost her licence for six months for AI citations nobody checked plus lying about it afterward — reported as the first Canadian case of a licence suspension, not just a costs order, tied to AI misuse. Both sit inside a fast-growing pattern: courtready.ca's Tom Macintosh Zheng counted 74 Canadian court and tribunal decisions involving AI misuse in the first six months of 2026 alone, against 87 in all of 2025 and 7 in 2024, with roughly 81% of decisions tracked since 2024 involving self-represented parties and 19% (32 decisions) involving lawyers who were sanctioned. Sources: Law Times News, “‘Out of Character’: Tribunal Explains Six-Month Suspension of Lawyer Who Lied About ChatGPT Use,” July 2026, and “Lawyer Who Used AI-Fabricated Citations Hit With $31,150 in Costs to LSO,” June 2026; Law Gazette (UK), July 2026; Law Society Tribunal case record, 2026 ONLSTH 136.
Before your next filing
Whatever tool touched a citation in your next factum, motion, or Tribunal submission, three checks take less time than a costs order:
- Open every case cited and confirm it exists on CanLII or your firm's primary research platform — not just that a hyperlink resolves somewhere.
- Confirm the case actually says what the citation claims, not just that the case is real.
- Know whether your research tool answers from a named, current corpus or from open-ended model recall — and treat any answer from the latter as unverified until you check it yourself.
Not ready to subscribe? Join the DROZlegal waitlist instead.
- Related: Is AI Legal Research Privileged in Canada?
- Related: Fabricated AI Citations in Canadian Courts: The full 2024–2026 case tracker, with year-by-year counts and the self-represented-vs-lawyer split