AI triples self-filed lawsuits but doesn't help win cases, study finds

A groundbreaking study from MIT and the University of Southern California, examining 4.5 million federal civil cases from 2005 to 2026, reveals that AI-generated lawsuits are flooding U.S. courts. The share of filings by self-represented litigants jumped from 11% in 2022 to 16.8% in 2025, with AI-text detector Pangram flagging 18% of filings as AI-written in 2026, up from just 1% in 2023. Judge Maritza Braswell of Colorado, a tech-savvy magistrate, attributes this surge to AI tools like ChatGPT and Microsoft Copilot. While she sees better-drafted pleadings, she also encounters hallucinated cases and fabricated quotes, requiring careful vetting. The clearer filings help her understand arguments faster, but the study confirms that AI assistance does not improve litigants' chances of winning—pro se plaintiffs still lose far more often than those with lawyers.

Beyond drafting, courts are now grappling with deeper legal questions. In Vermont, a viral Reddit post advocating use of Microsoft Copilot to draft writs of mandamus for immigration cases caused filings to skyrocket from 45 per year before 2022 to over 1,100 in 2024. Judges like William Garfinkel in Connecticut are questioning whether chatbot conversations dispensing legal advice should enjoy attorney-client privilege. A Michigan federal court recently began exploring this issue. Lawmakers are starting to debate who bears liability when AI gives bad legal advice. The study’s authors note that mounting a lawsuit involves complex tasks beyond text generation—including strategy, evidence gathering, and procedure—which AI still cannot fully handle. The result: courts get clearer paperwork, but equal justice remains elusive.