US courts struggle to manage surge in AI-generated lawsuits
While AI-assisted filings are clearer and easier for judges to decipher, they frequently contain errors and hallucinations, leaving litigants no more likely to win than those without legal representation.
US federal courts are grappling with a significant increase in lawsuits filed by self-represented litigants utilising artificial intelligence, with filings rising from 11 per cent of federal civil cases in 2022 to 16.8 per cent in 2025. A study by researchers from MIT and the University of Southern California (USC) found that the proportion of documents flagged as AI-generated rose from 1 per cent in 2023 to 18 per cent in 2026. The research, which examined 4.5 million federal civil cases, indicates that the volume of filings made by individuals without lawyers more than doubled compared to pre-2023 levels.
Judge Maritza Braswell, a federal magistrate judge in Colorado, attributes this surge to the growing accessibility of large language models. She notes that while AI-assisted filings are often clearer and easier to understand than handwritten submissions, they frequently contain hallucinations and fabricated quotes. Despite these errors, judges report that the improved clarity helps them better articulate arguments and rule on cases more efficiently, reducing the time previously spent deciphering cryptic, self-written documents.
However, the study by MIT’s Anand Shah and USC’s Joshua Levy found that litigants using AI remain significantly less likely to win their cases than those with legal representation. The researchers tested whether AI was driving the increase by running 1,600 randomly sampled court documents through Pangram, a commercial AI-text detector. The results confirmed a sharp rise in AI-generated content, yet the underlying complexity of mounting a lawsuit means that drafting text alone does not guarantee success.
Federal courts are currently divided on whether conversations between litigants and chatbots should be protected by attorney-client privilege or work-product doctrine. In February, a federal court in Michigan ruled that such conversations were shielded, while a New York court held the opposite, citing the lack of a reasonable expectation of confidentiality with AI companies. Judge Braswell has also ruled that such use should remain off-limits, noting that AI systems collect user data for training purposes, which complicates privacy expectations.
Lawmakers and courts are also beginning to address liability issues, including a pending lawsuit where Nippon Life Insurance Company alleges OpenAI’s ChatGPT practiced law without a licence and facilitated frivolous filings. OpenAI has argued that ChatGPT is not a person and does not possess legal knowledge, seeking dismissal of the case. Meanwhile, states and Congress are considering legislation to hold AI companies liable for providing bad legal advice and to ban chatbots from impersonating licensed professionals.
