Our Take
AI is making bad cases articulate, not good—judges read them faster but plaintiffs still lose more often than lawyers win for them.
Why it matters
Courts are drowning in AI-assisted filings from people who cannot afford counsel. The real problem isn't clarity; it's that chatbots offer false confidence in cases that were never winnable.
Do this week
In-house counsel: audit your litigation workflow for AI-generated opponent filings by end of week so you can flag hallucinations and fabricated citations before they muddy settlement talks.
Self-represented lawsuits spike, fueled by AI drafting tools
Federal magistrate courts are processing a surge of filings from people without lawyers, and AI is the catalyst. A study of 4.5 million federal civil cases from 2005 to 2026 found that self-represented litigants increased their share from 11% in 2022 to 16.8% in 2025. Within those cases, filing volume more than doubled from pre-2023 levels (per MIT and USC researchers Anand Shah and Joshua Levy).
When the researchers ran 1,600 randomly sampled court documents through Pangram, a commercial AI-text detector, the share flagged as AI-generated rose from 1% in 2023 to 18% in 2026. Judge Maritza Braswell, a federal magistrate in Colorado, confirms the pattern: she recognizes LLM prose by its syntactic fingerprints and says she "sees AI use" in her chambers daily. She also spots hallucinated cases and fabricated quotes.
Online communities now trade DIY guides for suing the government via AI. A December 2024 Reddit post instructed immigration applicants to draft writs of mandamus with Microsoft Copilot, hire a lawyer for $150 to polish them, and file in Vermont's expedient federal district. Cases filed without lawyers in Vermont jumped from roughly 45 per year before 2022 to more than 1,100 in 2024.
Clearer pleadings don't mean better outcomes
Judge Braswell and other jurists report that AI-drafted motions are faster to parse than handwritten scrawls. She churns through AI-assisted filings quicker than human-written ones. "If I understand an argument a little bit better, I'm probably going to be able to help a little bit more," she says.
But access to legal prose is not access to justice. The study found that self-represented litigants lose more often than those with lawyers, and that gap persists even with AI help. As Levy noted, "mounting a lawsuit is a complex, multifaceted task. Not all of it is just drafting text." A person asking ChatGPT what their slip-and-fall case is worth may be told $700,000 when it is worth a fraction of that. Judge Allison Goddard in California has seen this play out: she called it "Dr. Google went to law school."
Courts are now fractured on whether conversations with chatbots should be privileged like attorney-client exchanges. In February, a Michigan federal court ruled such work product protected. Days later, a New York court held it was not, because Claude and ChatGPT are not attorneys and users have no reasonable expectation of privacy. Judge Braswell disagreed in March, noting that privacy expectations survive data-collection practices.
Liability is an open question. In March, Nippon Life Insurance sued OpenAI alleging ChatGPT practiced law without a license. OpenAI countered in May that ChatGPT is not a person and has no legal knowledge or skill. The case is pending. New York proposed legislation in March to ban chatbots from impersonating lawyers, even with user notification. Federal bills would do the same for doctors and other licensed professionals, but have not gained traction.
What courts and counsel should do
Judges handling pro se (self-represented) dockets should budget for manual spot-checks of AI-flagged filings. Judge Braswell screens documents herself, aware that hallucinations and invented citations hide inside otherwise coherent prose. A ruling that AI-drafted work product is sometimes privileged and sometimes not will create discovery disputes; establish clear local rules now.
Defense counsel should document every AI-generated filing on the record, including flagged passages and dates of detection. Opposing counsel will rely on chatbot errors (false case law, wrong statutory baselines, miscalculated damages). You benefit from contemporaneous written notice that you identified the error and flagged it before settlement.
Legislators should resist the impulse to ban chatbots from offering legal guidance. The real harm is not advice-giving; it is false confidence. Target liability instead: hold AI vendors accountable when their systems explicitly claim to draft legal documents or provide case valuations and those outputs materially deviate from applicable law or custom. Banning the label does not stop the use.