In the spring of 2026, a federal district court clerk in the Eastern District of Texas opened a civil complaint that began with a surprisingly eloquent recitation of constitutional grievances. The pleading cited obscure precedents, invoked the Due Process Clause with theatrical flair, and demanded $47 million in damages for a dispute that, on closer inspection, appeared to involve a man, his neighbor, and a fence that had been in the wrong place since 1987. The complaint was 34 pages long. It was also, in the judgment of the court's staff attorney, "almost entirely fabricated by a large language model."
This is not an isolated incident. It is the new normal.
According to a sweeping study published by researchers who analyzed 4.5 million civil lawsuits filed between 2005 and 2026 — encompassing 46 million docket entries — approximately one in five complaints now contains text generated by artificial intelligence. In 2023, that figure was roughly 1%. By 2026, it has surged to 18%. The trend line is not merely steep; it is vertical. And the American legal system, already groaning under decades of underfunding, case backlogs, and procedural bloat, is now facing a deluge of machine-generated litigation that it is structurally unprepared to handle.
The ChatGPT Plaintiff
The rise of self-filed lawsuits is not, in itself, a new phenomenon. Pro se litigants — individuals who represent themselves in court — have always existed at the margins of the civil justice system. What has changed is the volume, the polish, and the absurdity. Free AI tools like ChatGPT, Claude, and a growing ecosystem of legal-adjacent chatbots have democratized access to legal language without democratizing access to legal reasoning. The result is a generation of plaintiffs who can draft a complaint that looks like it was written by a $800-an-hour partner at Sullivan & Cromwell, but whose legal arguments collapse under the gentlest scrutiny.
"The courts take all filings seriously," one paralegal at a federal district court told MIT Technology Review in a recent investigation. "And all of this sh*t is clogging the system." That blunt assessment captures the operational reality. Federal judges are required by statute and ethical obligation to treat every complaint as a genuine petition for relief. They cannot simply dismiss a filing because it appears frivolous on its face — not without following due process, not without issuing an order to show cause, not without risking an appeal. And so every AI-generated complaint, however incoherent, however detached from legal reality, must be read, processed, docketed, and adjudicated. The administrative cost is staggering.
The complaints themselves run the gamut from the merely misguided to the legally hallucinatory. Some cite statutes that do not exist. Others name defendants who have no connection to the dispute. A surprising number include demands for relief that no court has the power to grant — one recent filing sought an injunction compelling the federal government to "restore the original Constitution of 1787," a document that, the plaintiff appeared not to realize, was never replaced. Another demanded "emotional damages from the algorithmic bias of the Google search bar." The language in each was fluent, confident, and utterly wrong.
From Access to Justice to Paperwork Nightmare
The irony is bitter. AI was supposed to solve the "justice gap" — the well-documented reality that the vast majority of Americans with a civil legal problem cannot afford representation. Legal aid organizations are chronically underfunded. Pro bono programs are overstretched. For the average person facing eviction, wrongful termination, or a predatory debt collection, the courthouse might as well be on Mars. AI tools were heralded as the great equalizer: a way for ordinary people to draft pleadings, understand their rights, and navigate a system that has always been stacked against the unrepresented.
Instead, the equalization has taken a perverse form. The tools have made it trivially easy to file anything, and the court system has no filter. The result is not justice. It is a paperwork nightmare.
"We are seeing a fundamental mismatch between the speed at which AI can generate legal documents and the speed at which the courts can process them," warned a researcher at MIT in an interview with The Decoder. "If this trend continues unabated, the federal courts could grind to a halt." That is not hyperbole. The federal judiciary processed approximately 360,000 civil cases in 2024. If 18% of new filings are AI-generated, and if even a fraction of those are frivolous or require extraordinary judicial attention to untangle, the cumulative drag on the system becomes existential. Clerks are already working overtime. Judges are delaying substantive hearings to wade through procedural motions generated by chatbots. The entire apparatus is slowing down.
Judges Fight Back — With Drastic Measures
Some judges have begun to fight back with what can only be described as drastic measures. In the Northern District of California, one magistrate judge now requires all pro se filers to certify under penalty of perjury that no portion of their submission was drafted with AI assistance. In the Southern District of New York, a district judge has begun using Pangram, an AI-content detection tool, to flag suspicious pleadings for enhanced review. The results are imperfect — Pangram and its competitors are known to produce false positives and false negatives — but the judge has described the approach as "better than nothing in a sea of noise."
Other judges have taken a more direct approach. Several have issued standing orders requiring attorneys (and, where applicable, pro se litigants) to disclose any use of generative AI in the preparation of court filings. Violations can result in sanctions, dismissal, or referral to disciplinary authorities. The American Bar Association is reportedly drafting model rules that would codify similar disclosure requirements nationwide. But enforcement is patchy, and the rules are reactive, not preventive. By the time a filing is flagged, it has already consumed court resources.
The problem is not limited to plaintiffs. In a twist that underscores the technology's ubiquity, at least two federal district court judges and one Georgia state judge have issued opinions that contained factual inaccuracies traced to AI-generated research. The judges had used AI tools to assist in drafting their opinions — a practice that is not prohibited in most jurisdictions — and the tools had "hallucinated" case citations, statutory interpretations, and historical facts. One federal opinion cited a Supreme Court precedent that does not exist. Another quoted a statutory provision that was repealed in 1978. The incidents have prompted calls for new rules governing judicial use of AI, but the larger lesson is clear: the technology is everywhere, and its errors are insidious precisely because they are wrapped in authoritative prose.
The Uncomfortable Math
The numbers do not lie. The research team that analyzed 4.5 million civil cases developed a novel detection methodology, combining stylometric analysis with large-language-model probability scoring, to identify text that was likely AI-generated. Their findings were stark: the inflection point came in late 2024, shortly after the release of more capable and widely accessible models. By early 2025, the rate of AI-drafted text in civil complaints had crossed 10%. By mid-2026, it hit 18%. Extrapolated forward, the researchers estimate that AI-generated content could account for the majority of new pro se filings by 2028.
The federal courts are not the only ones affected. State courts, which handle the overwhelming majority of civil cases in the United States, are reporting similar trends. Family courts are seeing AI-generated custody petitions that cite nonexistent psychological studies. Small claims courts are receiving AI-drafted demands for damages that bear no relationship to the actual disputes. Housing courts are overwhelmed with AI-generated eviction defenses that raise constitutional arguments wholly inapplicable to landlord-tenant proceedings. The pattern is the same everywhere: the language is fluent, the law is wrong, and the system must still respond.
Reuters reported in May 2026 that several federal circuit courts are now considering emergency rulemaking procedures to address the crisis. Proposals include mandatory AI-disclosure certificates, enhanced filing fees for pro se litigants, and even a temporary moratorium on electronic filing for certain categories of self-represented claims. None of these measures are without controversy. Civil liberties groups have raised concerns about access to justice. Tech companies have warned that detection tools are unreliable. And the courts themselves are divided on whether the problem is best addressed through rulemaking, legislation, or simply waiting for the technology to improve.
🔥 Hot Takes
1. The "Justice Gap" Was Always a Lie, and AI Just Exposed It. For decades, the legal profession has wrung its hands about access to justice while maintaining a guild system that makes legal representation unaffordable by design. AI didn't create the crisis — it revealed that the crisis was never about access to documents. It was about access to judgment, strategy, and power. Giving people a chatbot that can write a complaint in iambic pentameter doesn't fix the system. It just makes the system's dysfunction more visible, and more voluminous.
2. We Should Let the Courts Drown — It's the Only Way to Force Reform. The federal judiciary has been underfunded and understaffed for a generation. Congress treats the courts like a nuisance budget line. Clerks are overworked, judges are burning out, and case backlogs stretch for years. The AI lawsuit deluge is not the disease; it is the symptom that finally forces the medical examiners to look at the patient. If the system grinds to a halt, maybe — just maybe — lawmakers will be forced to fund it, modernize it, and admit that a 19th-century procedural apparatus cannot handle 21st-century disputes. Sometimes you have to let the house burn to get the fire department funded.
3. AI Detection in Courts Is a Privacy and Due Process Disaster Waiting to Happen. Judges are already experimenting with AI detectors like Pangram to flag suspicious filings. This is a terrible idea. These tools are probabilistic, not deterministic. They will flag legitimate filings by non-native English speakers, by people with neurodivergent writing styles, by anyone whose prose doesn't match the detector's training data. The result will be a two-tier system: one for people who write "normally," and one for people who get flagged, delayed, and scrutinized. That is not justice. That is algorithmic profiling in a courthouse, and it should be banned before it becomes standard practice.
4. The Real Scandal Is Judges Using AI to Write Opinions, Not Plaintiffs Using AI to File Complaints. Everyone is focused on the chatbot plaintiffs burying the system in garbage. But the far more dangerous problem is on the bench. When a federal judge uses AI to draft an opinion, and that opinion contains hallucinated precedents, the error becomes binding law for the parties before the court. It can be cited by other litigants. It can influence settlement negotiations. It can shape the development of doctrine. A bad complaint gets dismissed. A bad opinion becomes precedent. The judicial use of AI is the scandal that nobody wants to talk about because it implicates the people who are supposed to be the solution, not the problem.
The Bottom Line
The federal courts are being flooded with AI-generated lawsuits. The numbers are not in dispute: 18% of complaints, 4.5 million cases analyzed, 46 million docket entries reviewed. The trend is accelerating. The system is already overburdened. And the available responses — disclosure rules, detection tools, enhanced fees — are inadequate to the scale of the problem.
What happens next will depend on whether the legal system can adapt faster than the technology can proliferate. That is not a bet that history suggests making. The courts have been slow to adopt electronic filing, slow to embrace remote hearings, slow to modernize in virtually every respect. Now they are being asked to regulate a technology that changes faster than their rulemaking cycles. The mismatch is profound.
The justice gap remains real. Millions of Americans genuinely need legal help they cannot afford. AI could, in theory, help bridge that gap. But the bridge, as currently constructed, leads to a courthouse that is sinking under the weight of machine-generated paper. And until the system itself is rebuilt — not just patched, not just regulated, but fundamentally reimagined — the chatbot plaintiffs will keep filing, the judges will keep drowning, and the promise of democratized justice will remain exactly that: a promise, unfulfilled, buried under 34 pages of eloquent, confident, and utterly meaningless prose.