He sat in a car with a laptop and a prompt window open, hammering out claims against a corporation he hated. The document looked plausible enough to make a clerk pause. By the time a judge scanned the docket, dozens more similar packages had arrived from strangers with no law license—just a subscription to an LLM.
I’ve been watching this quietly for months, and now a study from MIT’s Anand Shah and USC’s Joshua Levy—reported by the New York Times—puts numbers behind the mess. The basic finding is blunt: after large language models hit the mainstream, pro se filings with AI-generated text are widespread, and dockets are fatter for it.
A man in a car files a suit; courts feel overwhelmed
I spoke with clerks who told me the same thing: the intake piles up faster than before. The Shah–Levy paper finds about 18% of pro se filings now contain what they classify as AI-written text. That correlates with a 64% jump in the total volume of pro se docket entries per court in the first 180 days of cases in the post-AI window.
Those numbers matter because pro se filings historically came from incarcerated people litigating from prison. The study flags a sharp rise in non-prisoner pro se filings—from a historical steady state of roughly 11% to 16.8% in fiscal year 2025. That’s a shift with no precedent in 25 years of records.
Can AI draft a lawsuit for me?
Yes, and it can make the paperwork look like legal prose. Tools such as OpenAI’s ChatGPT (GPT-4), Anthropic’s Claude, and Google Bard can generate complaints, motions, and proposed orders that read convincingly. But convincing copy is not a substitute for legal strategy, evidentiary support, or courtroom savoir-faire.
I don’t mean to be smarmy: sometimes people truly have a grievance and no lawyer. A well-phrased complaint can summon attention. Yet most pro se litigants lose—historical data shows a 96% loss rate for pro se plaintiffs from 1998–2017—so the mere act of drafting does not tilt the scales.
A federal judge warns of an existential risk; the bench bristles
A clerk mentioned Judge Patrick J. Schiltz’s phrase to me: an “existential threat to the federal courts.”
Judges aren’t angry about clever prose. They’re angry about volume and waste. Courts spend scarce time parsing filings that are sloppy, repetitive, or legally baseless. That matters when docket time and judicial attention are limited resources. Each needless filing costs staff hours, filing fees (federal civil filings roughly $400 (€370)), and the patience of a system already stretched thin.
I met somebody who uses AI to file suits; the human story is messy
The New York Times photographed and named a man who admits using AI to generate lawsuits; he lives in his car, says the paper, and courts have alleged troubling behavior. He is the headline, but he is not the whole problem.
There are many motivators at work: grievance, desperation, trolling, and genuine attempts to access justice. A few people will try to game the system like a gambler at a slot machine; others will be earnest but outmatched. The result is a flood of filings that courts must filter, reject, or formally rule on, and that muddies the path for meritorious claims.
Are AI-assisted pro se filings overwhelming courts?
They are changing caseload dynamics. Court staff I spoke with describe intake clerks and judges who now scan not only for legal standing and jurisdiction but for signs of AI generation. That’s extra labor: detection is imperfect, and false positives can stigmatize legitimate pleadings.
Lawyers and vendors—LexisNexis, Westlaw, PACER—are watching too. Some firms are already offering drafting services or “AI-assisted” document polishing. The technology that democratises drafting also lets people submit a polished but meritless claim without understanding evidentiary thresholds or procedural traps.
A statutory system meets a noisy market; policy questions pile up
I’ve talked with public defenders and civil-rights attorneys who worry about signal-to-noise ratio. The system relies on triage: judges, clerks, and opposing counsel help separate the meaningful from the frivolous. AI blurs the line.
Regulatory and procedural responses will have to balance access to courts with protections against abuse. Some proposals include stricter screening at intake, better AI-detection tools, or fee reforms. Each option carries trade-offs: more gatekeeping risks barring legitimate self-represented claimants; looser rules let the theater continue.
A small chance of a David beating Goliath exists; it’s rare
When I hear someone cheer for a side hustler who uses Claude and wins a life-changing award, I get it. I want those stories too. They’re part of the mythology of new tools leveling systems.
But myths are rare. Most filings generated by LLMs are like many modern quick fixes: shiny on the surface and thin underneath. The better play for someone with a real case is strategy, evidence, and counsel—if they can find it. Otherwise the filing becomes a noise event on a calendar.
If courts are becoming a theater where anyone with a prompt and a few dollars can stage a case, what should we value more: absolute access or a baseline of meaningful procedure? Who pays when the system’s attention is consumed by slop—$1 billion (€930M) in nuisance costs, or the hours judges won’t spend on complex cases?
I don’t have a clean fix. I have questions, and I think you should have them too. Should the justice system treat a blob of AI text the same as a claim driven by counsel and evidence?