Who's Submitting AI-Tainted Filings in Court? (cyberlaw.stanford.edu)

A new analysis of 114 U.S. court matters drawn from Damien Charlotin’s “AI Hallucination Cases” database (orders dated June 2023–Oct. 7, 2025) shows that filings tainted by AI-generated inaccuracies—fabricated cases, false quotes, or citations to overturned authority—are disproportionately coming from solo and small firms and from plaintiffs’ counsel. The reviewer identified 129 firms involved across the sample: 50.4% were solo practitioners and 39.5% were firms of 2–25 lawyers. Plaintiffs accounted for 56.1% of the incidents versus 30.7% for defendants. Only 30% of entries named the AI tool used; of those, half implicated some version of ChatGPT, followed by Westlaw, Anthropic’s Claude, Microsoft Copilot, Google Gemini, and LexisNexis tools. The findings matter because they show systemic risk: overstretched small practices are turning to generative AI for legal research and drafting, then relying on outputs that remain prone to hallucination—sometimes with sanctions, client harm, and additional court burden. The analysis echoes academic tests reporting high hallucination rates across major legal-AI offerings (GPT-4, Lexis+, Westlaw, Ask Practical Law AI). Practical takeaways: courts’ advisories and CLEs haven’t prevented repeat errors; firms must validate citations manually, regulators and vendors should improve disclosure and retrieval reliability, and the legal community needs stronger guardrails for trustworthy AI use in high-stakes advocacy.