Disclosure: DROZlegal publishes this tracker and builds a practice-automation product for Canadian law firms. Every figure below traces to Toronto lawyer Tom Macintosh Zheng's published CanLII research and live tracking database at courtready.ca, Canadian HR Reporter, and Law Times News — each named, dated, and linked — not to DROZlegal's own data. This tracker reflects a live, continuously updated dataset; where an earlier snapshot is cited elsewhere (including on this site), both dates are shown side by side below rather than treated as a single fixed count.
The pattern, year by year
The cleanest, most precisely dated cut of this data covers CanLII decisions from January 1, 2024 through March 26, 2026.
| Year | Decisions logged | What stands out |
|---|---|---|
| 2024 | 7 | First cases identified; still a rare, isolated event. |
| 2025 | 86 | More than 12× 2024's total in a single year — the same year as Ko v. Li, 2025 ONSC 2766, the Ontario Superior Court's first widely reported AI-fabricated-factum contempt referral. |
| 2026 (Jan 1–Mar 26) | 39 | Already 45% of 2025's full-year count in under three months — the same stretch that produced the Law Society Tribunal's $31,150 costs order in Mazaheri v. Law Society of Ontario, 2026 ONLSTH 112. |
| Total | 132 | Across 44 courts and tribunals in six provinces plus the federal jurisdiction. A court or tribunal found or presumed AI generated the fake citations in 96 of the 132 (73%). |
Source: Tom Macintosh Zheng (Courtready.ca), systematic CanLII review cited in connection with Mazaheri v. Law Society of Ontario, 2026 ONLSTH 112; reported by Canadian HR Reporter, June 18, 2026.
This is a snapshot, not a fixed historical record. Courtready.ca's underlying database updates on a rolling basis, and different articles — including two of ours — have cited it at different moments. Our own coverage of the Mazaheri decision cites a later checkpoint of this same tracker: 168 total decisions, split 7 (2024) / 87 (2025) / 74 (first half of 2026 alone). The bigger 2026 figure there isn't a contradiction — its cutoff (June 30) simply runs three months later than the 132-count's March 26 cutoff above. By September 1, 2026, courtready.ca's live database showed the total had climbed again, to 252 decisions across 61 courts and tribunals (210 of them specifically involving fictitious case citations; the rest, invented legal rules or fabricated quotes). Whichever checkpoint you cite, the direction is the same: this keeps getting worse, faster than most firms' internal review habits are keeping pace.
Who's actually filing the fake citations
Self-represented litigants file the large majority of these — but "the large majority" has held remarkably steady even as the total nearly tripled.
| Snapshot | Total decisions tracked | Self-represented litigants | Represented (lawyer-assisted) parties |
|---|---|---|---|
| Reported alongside the Mazaheri decision, June 2026 | 168 (51 courts/tribunals) | 81% (136) | 19% (32) |
| Live tracker, updated September 1, 2026 | 252 (61 courts/tribunals) | 83% (208) | 17% (44) |
Source: Courtready.ca, "Fictitious Citations in Canadian Courts" (accessed September 4, 2026); June 2026 figures per Canadian HR Reporter, cited above.
That doesn't make this only a self-represented-litigant problem. Roughly one in every five to six of these decisions involves a lawyer's own filing — including the two Ontario cases below.
Two Ontario names behind the numbers
Shahryar Mazaheri — already suspended from practice — filed Law Society Tribunal motions relying on Grok-generated citations that either didn't exist or had nothing to do with his arguments. The Tribunal called it "an additional and significantly aggravating factor" and ordered him to pay $31,150 in costs, reported at the time as the largest AI-citation costs award by a Canadian court or tribunal. We cover the full story — including the Law Society of Ontario's own 2026 budget decision to help fund CanLII's verified-research tooling — in our companion post on the Mazaheri decision; this tracker won't repeat those facts.
The more striking throughline belongs to one lawyer, twice, a year apart. In Ko v. Li, 2025 ONSC 2766, an Ontario Superior Court judge found a family-law factum relied on case citations that didn't resolve to real decisions and ordered counsel — identified in that proceeding as Jisuh Lee — to show cause for contempt; she withdrew the factum and apologized. A little over a year later, the Law Society Tribunal suspended a lawyer's licence for six months and ordered $10,000 in costs (2026 ONLSTH 136, decision dated July 16, 2026) after finding she had relied on a ChatGPT-generated factum — again — in what began as a family-law matter and became an estate matter, then misrepresented her AI use to both the court and the Law Society before admitting it. The Tribunal's own licensee record names her "Mary (a.k.a. Jisuh) Hyun-Sook Lee." Same lawyer, same underlying failure, one year and one enforcement level apart: a contempt referral in 2025, a licence suspension in 2026.
Update, September 8, 2026. Every case on this page involves a fabricated citation — a case name, quotation, or legal proposition that doesn't exist. A related but separate problem has emerged alongside it: AI-generated material submitted directly as evidence, not as a citation. See our new tracker on AI-generated evidence in Canadian courts for seven named rulings on that distinct issue, plus British Columbia's Civil Resolution Tribunal rule banning AI-altered evidence outright — effective February 1, 2026, the first Canadian rule of its kind.
What to do with this if you're evaluating your own filings
If AI touches drafting anywhere in your firm's pipeline before something reaches a court or tribunal, three related resources go deeper than this tracker does. For what a verified-citation research architecture actually requires before a case name reaches a filing — and the questions worth putting to any vendor first — see our guide to AI legal research software for Canadian firms. If the fabricated citation would have landed inside a motion or factum specifically, our breakdown of AI motion and factum drafting software for Ontario litigators covers the citation-linking and lawyer-review step built to catch it before filing. And for the full Mazaheri story, including the regulator's own funding decision, see our companion post on that costs order — a related page tracking the same underlying dataset, not a second, independent count.
Not ready to subscribe? Join the DROZlegal waitlist instead.