Legal press documents exponential growth in AI-hallucinated court filings

By late December 2025, legal-press coverage was treating AI-fabricated citations in court filings as a tracked, escalating problem rather than a series of isolated embarrassments. Bloomberg Law reported that a public database maintained by researcher Damien Charlotin had logged roughly 712 court decisions worldwide in which a judge found or strongly suspected that a party had relied on AI-hallucinated case law — about 90% of them from 2025 alone.

Charlotin, who built the database by hand from court filings and legal-press reports, told Bloomberg Law the rate of new entries had itself accelerated: he had been adding two or three cases a day as recently as September, rising to five or six a day by December. Riana Pfefferkorn, a policy fellow at Stanford’s Institute for Human-Centered AI, was quoted calling the trend “metastasizing in size” and predicting it would come to be treated as “a core problem” rather than a novelty. The article documented monetary sanctions in the cases it surveyed ranging from roughly $1,500 to $59,500, and judges in several jurisdictions describing the time spent identifying fabricated citations as a drain on already-stretched court resources.

The figures came with an acknowledged undercount built in: the database only captures instances a court explicitly flagged in a written decision, so cases caught informally, settled before judgment, or heard in state trial courts without searchable online records fall outside it. What the count did establish was direction and pace rather than a precise total — a shift from Mata v. Avianca and similar early cases being treated as one-off cautionary tales toward a body of jurisprudence large enough, and growing fast enough, for judges, bar associations and legal-technology vendors to treat as a standing feature of practice rather than an anomaly.