Database documents 120 court cases with AI-generated false citations
A French lawyer created a database tracking AI hallucinations in legal proceedings, documenting 120 cases where lawyers used generative AI that fabricated citations to non-existent court decisions. More than 20 cases were caught in the month before the database was publicized in May 2025, with the rate of discovered incidents increasing throughout the year.
What happened
A database created by French lawyer and data scientist Damien Charlotin in early May 2025 documented 120 instances of AI hallucinations in court cases dating back to June 2023. The database tracks cases where AI chatbots generated fake citations, typically to non-existent court decisions used to establish legal precedent.
More than 20 legal professionals were caught using AI hallucinations in court filings in the month before the database was publicized. Lawyers representing Anthropic used a false attribution generated by Claude in expert testimony, and law firm Levidow, Levidow & Oberman was caught in 2023 citing fake cases generated by ChatGPT in the Mata v. Avianca case.
Reported impact
The increasing use of generative AI to automate legal research and writing has made hallucinated citations easier to produce and harder to catch. Judges and opposing lawyers must now check citations for AI hallucinations in addition to verifying accuracy against real court decisions.
Index analysis
The database found 36 instances of AI hallucinations in court cases during 2024, the first full year of tracking, which jumped to 48 in 2025 with the year only half complete. Charlotin noted that the tracked cases represent only those caught in discovery, and the actual universe of hallucinated content in legal documents is necessarily wider.
Governance controls
The tracking of AI hallucinations in legal proceedings reveals gaps in oversight of generative AI use by legal professionals. The increasing prevalence of discovered cases suggests insufficient protocols exist to detect and prevent the use of AI-generated false citations before filing.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.