Since our June 4 coverage of the doubling of pro se filings since 2023, a deeper look at the underlying study reveals numbers that reframe the entire debate.

The Actual Data

Researchers Anand Shah at MIT and Joshua Levy at the University of Southern California analyzed 4.5 million federal civil cases spanning 2005 to 2026. Their findings, reported by MIT Technology Review, are stark.

The share of lawsuits filed by self-represented people climbed from 11% in 2022 to 16.8% in 2025. Within that group, the raw volume of filings more than doubled from pre-2023 levels.

To test whether AI was actually driving the surge, Shah and Levy ran 1,600 randomly sampled court documents through Pangram, a commercial AI-text detection tool. The share flagged as AI-generated went from 1% in 2023 to 18% in 2026.

What Judges Are Actually Seeing

Federal magistrate Judge Maritza Braswell in Colorado is processing this reality every day. According to MIT Technology Review, she says the volume of pro se filings in her chambers has noticeably jumped — and she directly attributes it to AI.

Braswell isn't purely alarmed. "I'm also actually seeing better-drafted pleadings," she told MIT Technology Review.

Court documents filed without lawyers have historically ranged from barely legible to completely incoherent. Judges are legally required to read them charitably — a time-consuming, often frustrating obligation. AI is, at minimum, making that job faster.

Braswell has flagged hallucinated cases and fabricated quotes buried in otherwise clean-looking documents. The prose looks professional. The citations sometimes don't exist.

The Access Question

AI is giving poor people a ticket into the courthouse. The legal system has always been pay-to-play — if you can't afford an attorney, your odds of navigating federal civil procedure on your own are roughly zero.

Access alone isn't solving the underlying problem. The MIT/USC study found that expanded filing rates are NOT translating into wins. More people are getting in the door. They're still losing.

People are spending more time, more emotional energy, and potentially more money on cases they were going to lose anyway — just with better-formatted paperwork.

The Liability Vacuum

If a licensed attorney gives you bad advice and you lose your case, they can face bar discipline, malpractice suits, and sanctions. There is a defined accountability chain.

If ChatGPT tells you that a legal theory exists, you file on it, the case gets tossed because the cited precedent is fabricated, and you lose — who pays? Not OpenAI. Not Google. Not Anthropic.

According to MIT Technology Review, a growing number of state lawmakers are starting to ask this question. Nobody has answered it yet. There is currently no federal framework governing chatbot liability for legal advice.

The AI companies' terms of service explicitly disclaim liability for professional advice. They're legally protected by those disclaimers. The person who trusted the bot is not.

What Coverage Is Getting Wrong

Left-leaning coverage — including the MIT Technology Review framing — tends to emphasize the access-to-justice angle as an unambiguous positive. More filings, more democracy, more people having their day in court.

If AI is flooding courts with unwinnable cases, it's consuming judicial resources that could go toward cases with merit. It may also be giving vulnerable people false hope — burning months of their lives on litigation doomed from the start because a chatbot said they had a strong case.

Right-leaning coverage, where it exists on this topic, tends to frame AI hallucinations purely as a Big Tech accountability story. The hallucination problem is real but secondary. The structural issue — no liability framework, no consumer protection, no professional standard — is what actually needs solving.

What Needs to Happen

Three things, none of which are happening fast enough.

First, Congress or the FTC needs to define what legal information versus legal advice means when it comes from an AI — and attach liability to the latter. Human lawyers have that distinction. Chatbots don't.

Second, courts need consistent disclosure rules. Judge Braswell has developed her own AI-detection instincts. The judge three courtrooms over may have none. A patchwork of individual judicial approaches creates inconsistency across the system.

Third, someone needs to study whether AI-assisted pro se litigants are winning at higher rates than unassisted ones. The MIT/USC study shows more filings. It doesn't yet tell us whether outcomes are improving. That data would settle a lot of arguments.

Until then, AI is democratizing access to a legal process that still grinds most self-represented people into dust — and the companies profiting from that are bearing exactly none of the consequences.