The number of artificial intelligence-generated errors, commonly referred to as "hallucinations," in U.S. court filings declined during the second and third quarters of 2026. Data from legal researchers indicates that while the prevalence of made-up or misleading citations has slowed since a peak in the spring, federal judges continue to issue warnings regarding the volume of inaccurate AI-assisted submissions.
The decline follows a surge in AI-related legal errors that began in 2022, corresponding with the mass-market debut of generative AI technology. Legal scholar Damien Charlotin, a senior researcher at HEC Paris, has tracked at least 1,500 such cases in U.S. state and federal courts through a dedicated database. Charlotin attributed the recent slowdown to improved large language models (LLMs) and increased caution by users, though he noted the trend might also reflect fewer errors being reported when they are deemed minor.
On Thursday, October 22, 2026, U.S. District Judge John Tuchi in Arizona highlighted the ongoing issue while dismissing a civil rights lawsuit that appeared to be produced using AI. Tuchi stated that for every effective AI-assisted filing, the court receives ten "obstructive" ones. The judge permitted the plaintiff, who was not represented by an attorney, to file an amended case but expressed concern over the "sheer verbiage" and volume of AI submissions currently affecting the court system.
Legal professionals are also facing scrutiny, as lawyers remain ethically bound to verify the accuracy of every submission. Recent incidents include a Florida lawyer referred for possible discipline on October 21, 2026, and the removal of a judge from a case involving a DEI ban on October 22 after AI errors were identified. These incidents highlight the risk of sanctions if the resulting "hallucinations"—such as false case citations—are not corrected before filing.
In the long term, Charlotin suggest that as AI tools reach a 99% accuracy rate, lawyers may lower their guard and stop checking outputs, potentially leading to future errors. The current decline in reported hallucinations marks a period where newer models are being tested against judicial oversight. The next steps for the court system involve ongoing monitoring by judges, who have not yet barred the technology but are increasingly using orders and sanctions to manage the volume of AI-generated content. No specific deadline for new federal AI regulations was reported.