US federal courts are absorbing a surge of AI-drafted lawsuits from people without lawyers, and judges are starting to set the ground rules for what a chatbot can and cannot do inside a case. A new study of 4.5 million federal civil cases from 2005 to 2026 by Anand Shah of MIT and Joshua Levy of the University of Southern California found that the share of lawsuits brought by self-represented people rose from 11% in 2022 to 16.8% in 2025, with the number of filings more than doubling from pre-2023 levels. The researchers ran 1,600 randomly sampled court documents through commercial detector Pangram and found AI-generated writing in 1% of filings in 2023, climbing to 18% by 2026.
Federal Magistrate Judge Maritza Braswell in Colorado attributes much of the increase to people using ChatGPT, Claude, and Gemini to draft pleadings they could not otherwise afford to write. Braswell, who uses AI to vet documents herself, says she now recognizes the prose style of large language models, along with the occasional hallucinated citation and fabricated quote. She is moving through AI-assisted motions faster than handwritten ones, even though wider access has not improved win rates for self-represented litigants.
The cost structure driving the surge is now visible online. A viral Reddit post from December 2024 walked immigration applicants through suing United States Citizenship and Immigration Services over delayed reviews: draft a writ of mandamus with Microsoft Copilot, pay a lawyer $150 to polish it, and file in the District of Vermont. Cases filed by people without lawyers in Vermont rose from roughly 45 a year before 2022 to more than 1,100 in 2024.
Key facts
- 01Self-represented lawsuits rose from 11% of US federal civil cases in 2022 to 16.8% in 2025, with filings more than doubling pre-2023 levels.
- 02A sample of 1,600 court documents run through Pangram showed AI-written share climbing from 1% in 2023 to 18% in 2026.
- 03Vermont saw pro se filings against USCIS jump from roughly 45 a year before 2022 to more than 1,100 in 2024.
- 04Nippon Life Insurance Company sued OpenAI in March 2026 alleging ChatGPT practiced law without a license; OpenAI moved to dismiss in May.
- 05Federal courts split in February 2026: Michigan ruled ChatGPT prep was protected work product; New York ruled Claude chats were not privileged.
Braswell says the clearer filings make her job easier, not harder, because she can finally understand what unrepresented litigants are arguing.
“I have to be really careful because some of them contain hallucinations and errors, but I can generally understand what they're arguing better with AI assistance from them than without it.”— Maritza Braswell, US Federal Magistrate Judge, Colorado
Whether AI changes outcomes is a separate question. The Shah-Levy study found that pro se litigants remain far more likely to lose than represented ones, and AI tools have not shifted that gap. Drafting a competent motion is only one step in litigation; discovery, evidence handling, and courtroom procedure remain.
Levy, one of the study's co-authors, frames the limit of chatbot-assisted lawyering bluntly in describing what AI cannot do for an unrepresented party.
“It turns out that mounting a lawsuit is a complex, multifaceted task. Not all of it is just drafting text.”— Joshua Levy, University of Southern California researcher
A second front has opened over whether chatbot conversations carry any legal protection. In February 2026, a federal court in Michigan ruled that a self-represented litigant's ChatGPT exchanges used to prepare her case qualified as work product, shielding them from the opposing side. The same day, a federal court in New York reached the opposite conclusion for a criminal defendant's Claude conversations, holding that Claude is not an attorney and that AI companies' ability to disclose user data to third parties defeats any reasonable expectation of confidentiality.
In March, Judge Braswell ruled the other way, writing that data collection by ChatGPT, Claude, Gemini, and others does not eliminate all expectations of privacy. Federal Magistrate Judge William Garfinkel in Connecticut, who has served three decades on the bench, says there is a credible argument that conversations with Claude, ChatGPT, or Grok should carry some protection. Courts remain split with no appellate ruling to settle the question.
Liability is the third unresolved issue. Federal Magistrate Judge Allison Goddard in California has watched self-represented plaintiffs arrive at settlement talks with inflated demands generated by ChatGPT; in one slip-and-fall case, a plaintiff asked for $700,000, well above what the case was worth. Goddard calls it Dr. Google going to law school. In March 2026, Nippon Life Insurance Company sued OpenAI, alleging ChatGPT practiced law without a license by helping a woman reopen a settled case and flooding the court with frivolous filings. OpenAI moved to dismiss in May.
The case is still pending. State and federal lawmakers are now drafting around the same problem: New York introduced a bill in March that would bar chatbots from impersonating lawyers even with a disclaimer, and Congress is weighing bills to prohibit chatbots from posing as lawyers, doctors, and other licensed professionals. None of the federal bills has gained traction.
For the AI industry, the courtroom data is a leading indicator of where consumer use of general-purpose chatbots is heading. ChatGPT, Claude, and Gemini are already standing in for paid professionals at the low end of legal services, where the alternative for most users is nothing at all. The unresolved questions over privilege, work product, and unauthorized-practice claims will shape how aggressively OpenAI, Anthropic, Google, and xAI can market models for legal, medical, and financial use without taking on duties their terms of service explicitly disclaim. The Nippon case is the first real test of whether a court will accept the position OpenAI argued in its May filing — that ChatGPT is a tool, not a practitioner — and the answer will set the template for every chatbot vendor whose product ends up cited inside a complaint.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
