AI Regulation Tracker  /  Court ruling

Canada Orders $31,150 Cost Sanction Over a Lawyer's AI-Fabricated Citations

An Ontario law society tribunal imposed what is reported as the largest AI-related cost penalty in Canada after a lawyer filed submissions, generated with Grok, that cited cases which did not exist. An Alberta appeal ruling adds that the duty to check citations cannot be handed off to staff, contractors, or an internal AI ban.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

A Canadian legal regulator has set a new marker for how expensive an artificial intelligence mistake can be. In Mazaheri v Law Society of Ontario, 2026 ONLSTH 112, the Law Society Tribunal ordered a lawyer to pay $31,150 in costs after filing submissions that cited cases which did not exist. The submissions had been produced with the AI tool Grok. One Toronto lawyer described the figure as the largest AI-related cost award in Canada by far, and it lands at a moment when courts across the country are losing patience with fabricated citations. The tribunal counted 132 cases involving fabricated citations on CanLII, 7 in 2024, 86 in 2025, and 39 in early 2026.

What the tribunal decided

The ruling came from the Tribunal's Hearing Division on June 12, 2026, in File 25H-139. The core problem was not that the lawyer used an AI tool. It was that the resulting authorities were never checked against a real reporter or database before they went into a filing. When a court or tribunal cannot find the cases a party relies on, the party has wasted the other side's time and the adjudicator's, and Canadian tribunals increasingly answer that waste with a personal cost award against the lawyer rather than the client.

For a working professional, the number matters as much as the principle. A five-figure personal cost order is not a slap on the wrist. It signals that Canadian adjudicators are prepared to price the harm of unverified AI output at a level meant to deter, and to attach the bill to the individual who signed. Costs on that scale also travel. A personal award, and the regulatory record that comes with it, is the kind of fact that surfaces in future retainer discussions, insurance renewals, and opposing counsel's research. The reputational tail can outlast the payment.

The tribunal's framing is worth keeping in mind precisely because it is ordinary. The lawyer was not accused of inventing cases on purpose. The failure was one of process: trusting output that read like law without confirming it was law. That is exactly the failure most likely to happen to a careful practitioner under deadline pressure, which is why the ruling reads less like an outlier and more like a warning aimed at the whole profession.

The Alberta ruling closes the escape hatches

The second development removes the excuses a firm might reach for. In Reddy v Saroya, 2026 ABCA 20, the Alberta Court of Appeal addressed a factum drafted by a third-party contractor that contained hallucinated, non-existent case law. The lawyer whose name was on the filing, Christopher Souster of Nimmons Law Office, had prohibited lawyers and staff at his firm from using generative AI since November 2024, consistent with Law Society of Alberta guidance. It did not matter. The court held that the lawyer whose name appears on the filed document bears ultimate responsibility for its contents, even though a contractor prepared it and may have concealed the use of AI. Souster was ordered to pay $17,550 plus GST personally.

That is the part professionals outside Canada should read twice. An internal ban on AI, on its own, is not a defense. If a contractor two steps removed from the file uses a chatbot and the bad citation ends up in your brief, the signing lawyer still owns it. Responsibility follows the signature, not the org chart.

The duty attaches at the moment of filing

Read together, the two decisions describe a consistent Canadian standard. The verification duty attaches when a lawyer signs and files. Using generative AI without personally checking every citation is sanctionable, can trigger a personal cost award, and can lead to a law society or regulatory referral. The obligation is not satisfied by a subordinate's assurance that the research was done, nor by a tool's confident tone.

What it does not do

These rulings do not prohibit AI. Neither decision bars lawyers from using Grok or any other tool to draft, summarize, or research, and neither treats AI use as misconduct in itself. What they penalize is filing unverified output. A lawyer who uses AI and then confirms each authority in a real database is not the target. The sanction is reserved for the failure to check.

For a US reader, the Canadian cases rhyme with the fabricated-citation sanctions already seen in a New York appellate matter and at the Ninth Circuit, but they add two distinctly Canadian features. First, the Mazaheri cost figure was described by a Toronto lawyer as the largest AI-related award in Canada by far. Second, the Alberta ruling places ultimate responsibility on the lawyer whose name is on the filing, so a firm cannot insulate itself through a contractor arrangement or an internal AI ban. US firms with Canadian operations, or with contractors filing north of the border, should assume the same standard applies to them.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.