When AI Floods Courts with Lawsuits From Unrepresented Citizens

Federal Judge Maritza Braswell, a magistrate in Colorado, spends a considerable portion of her workday reviewing judicial filings submitted by people without lawyers. Many of these documents now arrive drafted with generative AI tools. It's not always obvious at first glance, but patterns emerge: language more polished than usual, seemingly solid arguments that collapse when citations are verified, and references to case law that simply does not exist.

That is precisely what MIT Technology Review documents in its June 4 report: US courts are managing a growing volume of pro se litigation, with citizens representing themselves, amplified by AI tools that allow them to draft filings that previously required legal training. The phenomenon has two sides, and neither is straightforward.

What's happening in the courts

Pro se litigants are not new. People have always existed who, due to lack of financial resources or because their case is not profitable for a law firm, appear before a court without representation. What has changed is the quantity and appearance of the filings they submit. A large language model can produce in minutes a document with correct formatting, coherent legal terminology, and an argumentative structure that, on the surface, appears valid.

The problem arises when these models hallucinate. They cite decisions that don't exist, misinterpret applicable law, or construct arguments based on false premises. Judge Braswell and her colleagues must read each document carefully, aware that behind it may be a vulnerable person who has placed real hopes in an automatically generated and unverified text.

Courts are responding in different ways. Some judges have begun requiring sworn statements about AI use in preparing filings. Others apply additional review filters. No systematic solution exists yet.

Why it matters beyond the courts

This case is relevant for anyone working with general-purpose AI tools, including models in the Claude family. Not because the models are responsible for how they're used, but because it illustrates a pattern that recurs in different professional contexts: AI lowers the barrier to entry for tasks that previously required expertise, but it does not eliminate the need for that expertise to validate the result.

An engineer using Claude Code to generate code can review whether it compiles and passes tests. A citizen without legal training who uses a chatbot to draft a lawsuit does not have the tools to detect that the decision being cited was never issued. The asymmetry between ease of production and ability to validate is the core of the problem.

There is also a dimension of access to justice that deserves attention. Some who resort to these AI-generated filings do so because they cannot afford a lawyer. If the result is that their cases are dismissed due to factual or legal errors introduced by the model, the promise of democratizing access to professional tools becomes a trap.

Who should act and how

The MIT Tech Review report points to three actors: the courts themselves, which need clear and uniform policies on AI use in judicial filings; developers of general-purpose tools, who could incorporate more explicit warnings when their models are used in high-stakes contexts like legal or medical work; and legal aid organizations, which could explore how to use AI in a supervised way to genuinely help those without access to a lawyer.

Some states are already developing guidelines for pro se litigants using AI. It's a first step, though the speed of tool adoption clearly outpaces the speed of institutional response.

---

From our perspective, this case serves as a useful reminder that the utility of language models in high-consequence environments almost always depends on who oversees the output, not just who generates it. The gap between producing convincing text and producing correct text remains a human responsibility.