An AI hallucination in a court filing is a citation, quotation, or fact that a generative tool invented and a person then submitted as real. Damien Charlotin's public tracker listed 2,207 court decisions on hallucinated or allegedly hallucinated AI content as of October 23, 2026, including 1,504 from the United States and 234 from Canada. Court rules on both sides of the border place the duty to check on whoever signs the document, not on the software.
Key takeaways
- The tracker counts decisions, not filings, so the true number of fabricated citations is higher than 2,207.
- The Federal Court of Canada has asked for a declaration in the first paragraph of any filing that contains AI-generated content since its notice of May 7, 2024.
- Ontario requires a signed certification that cited authorities are authentic in factums filed on or after December 1, 2024.
- US Federal Rule of Civil Procedure 11 does not mention AI, but it requires a reasonable inquiry before anything is filed.
- A declaration of AI use does not draw an adverse inference in the Federal Court, while an unchecked fake citation can lead to costs or sanctions.
What the case count shows and what it leaves out
The Charlotin database, maintained by an individual researcher, includes decisions where a court or tribunal found or implied that a party relied on hallucinated material. The author adds that it also covers some decisions where AI use was alleged but not confirmed, and that it does not track the wider universe of fake citations that judges never noticed. The page states 2,207 cases identified so far, with 1,504 listed for the United States and 234 for Canada.
Two cautions apply. A decision appears only if a judge wrote about the problem, so quietly corrected filings never enter the count. And a decision counts once whether the error was one bad citation or a dozen. The figure is best read as evidence that the problem is widespread, not as a rate per lawyer or per filing.
Canadian rules: declaration and certification
The Federal Court issued its first notice on AI in court proceedings on December 20, 2023 and replaced it with an update dated May 7, 2024. Under the Federal Court notice, a document prepared for litigation that contains content created or generated by AI must say so in its first paragraph, naming either the whole document or the specific paragraphs. No declaration is needed where AI only suggested changes or critiqued text that a human wrote and then edited. The notice also says a declaration alone will not attract an adverse inference, and that the party signing a document remains responsible for its accuracy.
Ontario took a different route. Regulation 384/24 under the Courts of Justice Act, filed October 15, 2024, amended the Rules of Civil Procedure so that a factum must carry a statement, signed by the lawyer or the self-represented party, that the signer is satisfied the cited authorities are authentic. The Ontario Bar Association and Osler both report that the change came into force on December 1, 2024, and that authorities taken from government sites, CanLII, or court websites are presumed authentic. The certification applies whether or not AI was used.
The US position under Rule 11
Federal Rule of Civil Procedure 11(b) says that by presenting a paper to the court, an attorney or unrepresented party certifies, after an inquiry reasonable under the circumstances, that the legal contentions are warranted by existing law or by a nonfrivolous argument to change it. The rule does not name AI. Subsection 11(c)(4) lets a court impose nonmonetary directives as sanctions, limited to what suffices to deter repetition. The sources reviewed for this article did not show a single national AI disclosure rule for federal filings, so individual courts and judges should be checked case by case.
How the main rules compare
| Rule | What it asks for | AI named? | Status |
|---|---|---|---|
| Federal Court of Canada notice | Declaration in the first paragraph; human verification urged; use of recognized sources such as CanLII | Yes | In force, version of May 7, 2024 |
| Ontario Rules of Civil Procedure, O. Reg. 384/24 | Signed statement that cited authorities are authentic | No | In force since December 1, 2024 |
| Federal Rule of Civil Procedure 11 | Reasonable inquiry before signing; sanctions possible | No | In force; applies to any paper |
The pattern is consistent. Where AI is named, the rule asks for transparency. Where it is not, the rule asks for verification. Neither approach lets the tool carry the blame.
Consequences in practice
In Zhang v. Chen, 2024 BCSC 285, a British Columbia family-law lawyer put two cases generated by ChatGPT before the Supreme Court of British Columbia. According to a Canadian law-firm summary dated March 15, 2024, the court refused special costs, calling them an extraordinary remedy that requires reprehensible conduct or abuse of process, but made the lawyer personally liable for the opposing side's costs of four half-days. The result shows a middle path: no finding of bad faith, yet a real price for the error.
A verification checklist for filings
- Open every cited case on CanLII, a court website, or a commercial database and confirm that it exists.
- Match each quotation to the page or paragraph of the original decision.
- Confirm that the case actually says what the brief claims, not just that the name is real.
- Check the court's own notice or practice direction for a declaration or certification duty.
- Keep a record of which tool produced which paragraphs, so a declaration can be accurate.
This article is general information and not legal advice. Court rules change by court and by year, and anyone preparing a filing should read the current text of the applicable rule or consult a lawyer.
FAQ
Why do AI tools invent legal cases?
A language model predicts plausible text, and a case name with a believable court and year is plausible text. The model has no built-in step that checks a citation against a case database.
Does the Federal Court require a declaration for every use of AI?
No. The notice applies to content created or generated by AI in documents prepared for litigation. It excludes tools that only suggest or critique edits to human-written text.
Is a hallucinated citation always sanctioned?
No. Outcomes range from a warning to personal costs orders and other sanctions, depending on the court, the conduct, and whether the error was corrected quickly.
Where can a current count of cases be found?
The Charlotin database is updated as new decisions are found, so the totals above will be out of date soon after October 23, 2026.
Related reading
- hallucinations in image models
- AI ethics and responsibility
- contested AI policy questions
- AI's wider effects on society
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- AI Hallucination Cases database, Damien Charlotin, page updated October 23, 2026
- Notice to the Parties and the Profession: The Use of Artificial Intelligence in Court Proceedings, Federal Court of Canada, May 7, 2024
- Recent Amendments to the Rules of Civil Procedure, Ontario Bar Association, 2025
- Artificial advocacy: how Canadian courts and legislators are responding to generative AI, Osler, Hoskin and Harcourt LLP
- Rule 11, Federal Rules of Civil Procedure, Legal Information Institute, Cornell Law School
- Lawyer ordered to pay costs for citing fake cases in court material (Zhang v. Chen), Canadian law-firm summary, March 15, 2024