ChatGPT opens the courthouse door, and MIT sees the bill judges must pay

Generative AI has entered US federal courts through the most sensitive doorway: people who do not have a lawyer. According to The Decoder’s report, a new study from MIT and the University of Southern California shows that lawsuits filed without a lawyer have nearly doubled since ChatGPT became mainstream. More sharply, one in five complaints now contains AI-generated text.

That is the legal paradox in plain form. A system that was often unreachable because legal representation is expensive has become technically easier to enter. Someone who previously could not structure a complaint can now get a draft, a legal tone and the appearance of procedural form. But a court does not rule on tone. It has to assess jurisdiction, facts, claims, deadlines and applicable law. If AI produces text that sounds legal without carrying a viable legal case, the burden simply moves downstream to the court.

US federal courts already provide forms and guidance for people representing themselves, including official pro se forms. The problem is that general-purpose generative models do not behave like narrow procedural guides. They can expand claims, add broad legal phrasing and make a filing look more polished on the surface. For judges and court staff, that means more text to screen, more verification work and a higher risk that a genuinely serious case is buried among weak AI-assisted filings.

This is not just a story about people trying to game the courts with AI. The deeper problem is that AI can lower the barrier to entry while also degrading the quality of what enters. For people without lawyers, using a chatbot is a rational move. If the alternative is a blank page, the model looks like a practical translator between ordinary language and legal format. For courts, that same translator can create documents that look more coherent than they actually are.

That makes the issue social and regulatory before it is technical. Generative AI is not being measured here by model speed or context length. It is being measured by the amount of institutional labor it creates. If lawyerless filings have nearly doubled, courts are not merely receiving more cases. They are receiving more cases in which someone must separate meaningful access to justice from procedural noise.

The worst answer would be to close the gate automatically. Access to court should not depend on whether someone can afford a lawyer. But an equally bad answer is a court system where scarce resources are consumed by AI-drafted documents that fail basic legal requirements. The next phase will not be a simple ban-or-allow decision. It will be rule design: how to disclose AI assistance, how to filter filings without suppressing legitimate claims, and how to give self-represented people something better than a universal chatbot pretending to be a legal assistant.