- MIT and USC researchers analysed 4.5 million civil lawsuits and 46 million PACER entries; pro-se (self-represented) filings jumped from 11% (20-year baseline) to 16.8% in fiscal 2025.
- 1 in 5 complaints at US federal courts now contains AI-generated text.
- 59% of all growth in civil lawsuits in fiscal 2025 came from self-represented plaintiffs.
- The surge concentrates in formulaic case types (civil rights, consumer credit, foreclosures); patent and securities law show no effect.
What Happened
Lawsuits filed without a lawyer at US federal courts have nearly doubled since ChatGPT went mainstream, The Decoder reported, citing a new study from MIT and the University of Southern California. The researchers analysed 4.5 million civil lawsuits from fiscal years 2005 through 2026 and 46 million entries from the PACER electronic case registry. One in five complaints now contains AI-generated text, and judges are resorting to drastic measures to cope with the volume.
Why It Matters
The pro-se (self-represented) rate at US federal courts held steady at about 11% of all civil cases for two decades. In fiscal year 2025, it jumped to 16.8% — 41,490 pro-se filings, nearly double the pre-AI average. 59% of all growth in civil lawsuits in fiscal 2025 came from self-represented plaintiffs.
The finding is striking because US federal courts set the highest bar for unrepresented litigants in the American system. Filing fees run $405 — roughly twice what most state courts charge — and formal requirements for complaints are far stricter. Over 90% of all civil cases in the US go through state or local courts, where the AI effect is likely even larger, per the study.
Technical Details
The surge clusters in case types where formulaic document drafting does most of the work: civil rights complaints, consumer credit disputes, and foreclosures. Areas demanding sustained specialised knowledge — patent law, securities law — show no effect. The researchers read this as evidence that LLMs are cutting costs that were previously prohibitive for laypeople: drafting procedurally viable legal briefs.
The increase is almost entirely on the plaintiff side and shows up in 44 of 50 states at once, ruling out local explanations. Case durations and outcome distributions are largely unchanged, but activity within cases is exploding. The number of docket entries per court has grown substantially, creating administrative load on federal judges and clerks who must process the higher volume even when case-resolution rates remain stable.
Who’s Affected
US federal judges and court clerks face the immediate paperwork crisis. State courts — where 90%+ of civil litigation actually occurs — face a likely larger volume problem with weaker resources. Legal-aid organisations face a redefined role: the cost barrier they historically addressed has been partially solved by AI, but the resulting volume may overwhelm court capacity entirely. Legal-tech companies (Clio, Harvey, Legora) face renewed market opportunity for tools that help courts and clerks process AI-drafted filings efficiently. ChatGPT-equivalent providers face indirect regulatory scrutiny if the volume crisis prompts policy response.
What’s Next
The MIT-USC study is among the first quantitative measures of LLM impact on the US legal system. Expect follow-up research at the state-court level and on outcome quality (whether AI-drafted filings achieve outcomes comparable to lawyer-drafted ones). Court administrators may push for verification workflows analogous to arXiv’s one-year AI ban or biomedical reference-checking requirements. The broader “AI as legal infrastructure” question — covered by Anthropic’s Claude for Legal expansion and Harvey/Legora’s growth — gains a sharper empirical foundation.
