A new study of federal court filings suggests that artificial intelligence may be transforming one of the oldest — and most expensive — services in American society: representation in court.
The paper, now circulating for comments before publication, reports a nationwide surge in civil lawsuits filed by people acting pro se (as their own attorneys). The researchers conclude that many of these filings were researched, drafted, or otherwise heavily assisted by large language models (LLM) such as ChatGPT and Claude.
The study examined millions of federal court docket entries across multiple years and dozens of categories of litigation. Using large databases of filings, linguistic analysis, statistical comparisons with pre-ChatGPT baselines, and software to detect AI-generated language patterns, the researchers conclude that the rise of generative AI coincides with a major increase in pro se litigation in federal courts.
That increase is substantial. Federal civil filings surged after the public release of advanced LLMs. The overwhelming majority of the increase came from pro se plaintiffs. The authors report:
In the post-[AI] period, we find that the AI detection rate rises consistently, from near zero at the end of 2022 to 18 percent by early 2026, tracking the trajectory of LLM capability gains and diffusion of LLM adoption rather than any single event. By early 2026, roughly one in five complaint filings contains text that the detector classifies as AI-generated [emphasis added].
The right to represent oneself in federal court dates to the founding era and was codified by the First Congress in the Judiciary Act of 1789. In practice, however, the right often existed more in theory than reality. The Sixth Amendment and subsequent rulings assured a right to an attorney for those accused of a criminal offense, but no such protection exists in civil or administrative courts. Federal litigation is intimidating, technical, document-heavy, and expensive. Filing fees may be affordable, but legal representation often is not.
For generations, many Americans who believed they had been cheated, harassed, wrongly terminated, denied compensation, defamed, discriminated against, or otherwise mistreated under the law reached the same reluctant conclusion: it simply was not worth paying a lawyer and risking unpredictable legal costs to pursue a civil case. Artificial intelligence abruptly alters that calculation.
Chatbots Enter the Courtroom
A plaintiff can now ask ChatGPT to explain filing requirements, organize facts, summarize precedents, draft motions, prepare responses, and generate properly formatted legal documents at little or no cost. What once required thousands of dollars and weeks of expert legal work can be meaningfully attempted at home in an evening. (For what it’s worth, ChatGPT scored a 297 out of 400 on the Uniform Bar Exam, easily exceeding the passing score in nearly all US jurisdictions. Some 10 percent of US law school graduates never pass the exam.)
The jump in AI-assisted filings occurred across a wide range of legal categories, from real estate disputes and employment law to slander and libel claims, insurance matters, disability cases, and immigration proceedings. Of particular note, perhaps, is the rapid growth in filings related to detention and deportation, where detainees increasingly appear to be using AI tools to navigate the federal legal system without attorneys. Amid accelerating (often erratic) immigration enforcement, habeas corpus lawsuits have increased 8,400 percent. In other words, a market-generated AI technology may be broadening de facto access to constitutional protections for the groups least able to defend themselves within the legal system. From that perspective, this looks like an impressive democratization of access to justice.
For decades, reformers have argued that lower-income Americans effectively lack equal access to civil courts because professional legal services (“legal cognition”) are too expensive. Lawyers cost money because legal work requires highly trained cognitive labor: research, drafting, procedural knowledge, formatting, and endless paperwork. Legal aid programs and pro bono services help some people, but many are turned away because the basic economics remained unchanged.
Now, LLMs have dramatically reduced the cost of producing all that legal text and navigating legal procedure. The result may be a profound expansion of practical access to civil litigation in American history.
The Risks of AI Litigation
US courts evolved under the assumption that legal services would remain “scarce” and therefore expensive. That scarcity acted as a filtering mechanism. Lawyers didn’t merely draft documents. They discouraged weak claims, translated emotional grievances into legally actionable arguments, imposed professional discipline, and absorbed enormous procedural burdens before cases reached judges.
AI removes much of that friction.
The legal profession, for now, is deeply divided over the impact of what they call “robo-litigation” on the future of American justice. Some lawyers and judges see a coming wave of frivolous filings, fabricated citations, and procedural chaos. Others, including many organizations devoted to expanding access to legal services, see something closer to a breakthrough.
The American Bar Association recently conceded that generative AI could provide “affordable legal guidance” to people unable to hire attorneys but warned about inaccurate advice, ethical problems, and the unauthorized practice of law. Courts increasingly caution litigants that they remain fully responsible for errors generated by AI systems, including fabricated legal citations and imaginary precedents. Several lawyers already have been sanctioned after submitting briefs containing nonexistent cases hallucinated by AI tools.
Legal aid organizations are more enthusiastic than alarmed. The ABA’s own Center for Innovation has described AI as a potentially transformative tool for improving access to justice. The Thomson Reuters Institute recently called AI a “generational opportunity” to narrow the justice gap (unmet civil legal needs of Americans) by helping underserved populations navigate legal systems that historically have been too expensive and complex for ordinary citizens. One survey discussed by the Legal Services Corporation found that legal aid organizations are adopting AI tools at roughly twice the rate of the broader legal profession.
In short, the groups most directly confronted with unmet legal demand seem naturally eager to use AI. For legal aid attorneys and reform advocates, the advantages are obvious. A tenant facing eviction, a migrant contesting detention, a worker challenging dismissal, or a disabled claimant denied benefits suddenly can obtain procedural guidance, draft filings, organize evidence, and prepare legal arguments at little or no cost. What elite lawyers may view as a destabilizing automation can look different to people who previously had no practical access to legal representation at all.
Will Automation Overwhelm the System?
Critics reply that lowering the cost of legal access is also lowering the cost of producing legal “noise.” Federal judges complain of growing quantities of AI-assisted filings containing procedural errors, invented authorities, and poorly grounded legal arguments. Even accurate filings, however, increase pressure on courts that already struggle with overloaded dockets and limited judicial capacity. In many ways, the system’s stability relies on most people not exercising their rights. “Justice delayed” is one of the most politically sensitive measures of a court system.
And unlike private firms, the federal judiciary cannot rapidly gear up in response to surging demand. Adding more judges requires congressional action. Court procedures are governed by law, precedent, and elaborate procedural rules designed for consistency and due process — explicitly not for rapid adaptation.
In this sense, the federal courts may be confronting a classic problem identified decades ago by the economist Ludwig von Mises in Bureaucracy. Market institutions can respond dynamically to technological innovation because they are guided broadly by profit, loss, and competitive adaptation. Bureaucratic institutions operating under law necessarily move more cautiously because their authority depends upon rules, procedure, and limited discretion.
The courts now face precisely that asymmetry. The market has produced a revolutionary technology capable of dramatically lowering the cost of legal cognition. But the judicial system must respond through new legislation, revised procedures, modified administrative rules, and constitutional safeguards. Recently, for example, courts ruled that attorney-client privilege doesn’t apply to chatbots, even if the user is relying on them for legal counsel.
From Disruption to Democratization
Not so long ago, headlines about AI and law focused on whether ChatGPT could pass the bar exam and whether AI would threaten the jobs of junior associates drafting legal briefs.
Across profession after profession, AI is reducing the cost of cognitive labor once performed only by highly trained people. Journalism, accounting, customer service, education, finance, medicine, and software development — to name a few — all are confronting versions of the same disruption. Institutions built around expensive expertise suddenly face technologies capable of reproducing portions of that expertise at near-zero marginal cost. That we knew.
But few considered the more consequential finding that millions of ordinary citizens suddenly possess tools capable of helping them navigate the legal system. The relationship between ordinary Americans and the legal system is changing fundamentally. The constitutional right to represent oneself existed long before AI. What AI may now be doing is operationalizing that right on a mass scale for the first time in American history.
