For roughly twenty years, the share of self-represented plaintiffs in US federal civil cases sat at around 11% – stable, predictable, a known quantity that courts had built processes to manage. Then AI arrived, and between 2023 and early 2026, that figure jumped to nearly 17%. Docket activity in affected courts is up 158% against the pre-AI average. Pro se employment lawsuits rose 49% in 2025. Pro se Fair Housing Act filings were up 69% in the first nine months of the same year.
The data comes from a March 2026 paper by MIT researcher Anand V. Shah and USC’s Joshua Levy, who analysed 4.5 million federal civil lawsuits and 46 million PACER records from 2005 to 2026. Their central finding: AI has effectively lowered the barrier to filing a federal lawsuit to the point where the hardest part is no longer producing the paperwork. AI eliminated the document-generation bottleneck driving the recent surge in civil rights, consumer credit, and foreclosure cases.
Chief Judge Patrick Schiltz of the US District Court for Minnesota has called this trend an “existential threat” to federal courts. That’s strong language from a federal judge – it’s also accurate.
A Well-Formed Document Is Not A Valid Argument
The problem is best illustrated by a real case. A 69-year-old Minnesota resident named Donald Sove used ChatGPT and Claude to file a federal lawsuit against his ex-wife, her lawyer and the presiding judge. He submitted the complaint plus more than 50 supplementary documents and hundreds of pages of case law analysis. The case was dismissed – it didn’t contain a valid legal claim – but court clerks still had to read, record and make publicly available every single page. That’s the core problem: AI creates professional-looking legal documents at volume. It doesn’t create valid legal arguments.
At least 24 pro se litigants have received monetary sanctions since the second half of 2023 for litigating with AI. Federal Judge Virginia Kendall in Illinois fined a plaintiff $1,500 in March 2026 for submitting complaints containing fabricated case law – hallucinations dressed up as citations.
The asymmetry is brutal: filing costs for individual litigants have dropped dramatically while processing costs for courts and opposing parties have gone up.
More from Artificial Intelligence
The Disruption The Courts Didn’t Get
The Shah and Levy paper makes an essential, though difficult, truth: this is a one-sided technology shock. Litigants received powerful AI tools that can draft complaints, research precedents and format submissions; judges are left navigating the legal tech gap without any equivalent tools. The judicial system still runs on human judgment and administrative processing that wasn’t designed for this volume and has no obvious way to scale.
There’s no easy fix – as one researcher quoted in the study put it, “there is no easy margin along which to buy extra judge capacity.” You can’t solve a 158% increase in docket activity by hiring more clerks. The solution – if there’s one – is AI built for judges: grounded, auditable, designed for decision-making rather than document generation. That’s a whole different ballgame and a lot harder to build than the one the current wave of LegalTech tools is solving.
What This Means For LegalTech Founders
The lack of equal legal access is a reality – legal fees price most people out of the civil court system entirely, and AI has shifted that calculation for the first time in decades. Thomson Reuters’ AI for Justice programme, announced in September 2025, reported legal nonprofits serving up to 50% more clients daily using AI tools. Things are opening up to everyone, which is a good thing.
But the distance between ‘democratised access’ and ‘functional access’ is where the LegalTech opportunity actually exists. Right now, only 18% of small and solo law firms use AI-enabled tools, compared to over 70% of top commercial firms. The document generation problem is largely solved. The quality verification problem, making sure AI-assisted filings contain valid legal arguments rather than professionally formatted noise, is not. The navigation problem, helping someone who’s never filed a lawsuit understand what kind of claim they actually have before generating the documents, is not.
The founders building in this space face a choice between two very different products. The first is a tool that makes filing easier and faster – which is already in motion industry-wide, is generating the surge and is part of the problem. The second is a tool that helps people understand whether they have a viable case, what their realistic prospects are, and when the correct answer is ‘don’t file’. That second product is harder to build, harder to monetise and considerably more valuable to the justice system. It’s also the one that doesn’t contribute to a 158% increase in docket activity.
AI giving everyone a lawyer was always going to be complicated. The courts are finding out why.
