AI-fabricated content in court: 2,231 cases logged
A public database lists 2,231 court decisions that involve AI-made-up content as of 29 October 2026. Self-represented litigants account for 1,288 and lawyers for 880.
A public database now lists 2,231 court decisions that involve AI-made-up content, as of 29 October 2026. It lists 1,288 cases involving self-represented litigants and 880 involving lawyers [1].
The numbers
As of 29 October 2026, the database lists the following.
- Total cases: 2,231.
- Self-represented litigants: 1,288.
- Lawyers: 880.
- Judges: 32.
- Experts: 17.
- Other parties, such as prosecutors and paralegals: 12.
- By country, the United States has 1,514 cases and Canada 238.
The party counts add to 2,229, two fewer than the total, and the page does not explain the gap. Treat every count as approximate. The page was last updated on 28 October 2026.
What the database counts
Damien Charlotin, a researcher, maintains the database and licenses it under CC BY 4.0. It records court and tribunal decisions that address the use of AI in more than a passing reference, where the court found or implied that a party relied on hallucinated content. It also includes some decisions where AI use was alleged and not confirmed, which the author calls a judgment call. It does not track every fake citation or every use of AI in court filings. It is a researcher-run database, not a peer-reviewed study.
Our reading
Most of the cases come from people without a lawyer. That is expected, since they have no firm checking their work. The more useful signal is the other end of the table. 880 cases involve lawyers and 32 involve judges. Professionals with training and a duty of care also rely on made-up content.
The tool is not the common factor. The missing step is checking. A fabricated citation is cheap to catch, because someone opens the cited case and reads it. Each of these cases is one where that check did not happen or did not work.
For a law firm or a legal department, make verification a recorded step. Each cited authority gets opened, read and logged by a named person before the document leaves the building. For a board or general counsel, ask for the log, not for a promise that people are careful. For courts and regulators, count the cases that were caught at the checking stage, because those never reach this database and the reported cases are only the ones that failed.
The same rule holds outside law. Wherever AI produces something that people rely on, the work is the verification.
Verification is the job, not the tool.
Limits
The database counts reported decisions, so it shows where checking failed and tells us nothing about how often it worked. The database cannot separate growth in cases from growth in reporting. The party and country tables do not add exactly to the total.
Sources
- Charlotin, D., "AI Hallucination Cases database", https://www.damiencharlotin.com/hallucinations/ (page counts read and checked 2026-10-29; page last updated 28 October 2026)