AI Legal Advice Is Flooding the Courts
Federal judges across the US are confronting a sharp rise in AI legal advice showing up in self-filed lawsuits, according to reporting published June 4, 2026. The shift matters because clearer pleadings may improve access to court, but they also bring hallucinated citations, privacy disputes, and harder questions about liability when legal guidance comes from a chatbot. According to Technology Review’s report on AI-generated lawsuits, the trend is already visible in federal dockets and courtroom practice.
AI legal advice is showing up across federal court filings
The headline finding is straightforward: more people are filing cases without lawyers, and more of those filings appear to be AI-assisted. The study cited by Technology Review, by Anand Shah at MIT and Joshua Levy at USC, examined 4.5 million federal civil cases from 2005 to 2026 and found that the share of self-represented lawsuits rose from 11% in 2022 to 16.8% in 2025.
That increase is not just a volume story. In a sample of 1,600 court documents run through the commercial detector Pangram, the share flagged as containing AI-generated writing reportedly rose from 1% in 2023 to 18% in 2026. Judge Maritza Braswell, a federal magistrate judge in Colorado, told the publication she could often identify AI use by the prose style and by fabricated authorities, while also acknowledging that many pleadings are simply easier to read.
That distinction matters. Courts have long dealt with handwritten or poorly structured filings from people without counsel. If AI legal advice makes arguments more legible, judges can process them faster. But the operational trade-off is obvious: clearer language can hide weak legal reasoning, invented case law, or procedural errors.
Why AI makes lawsuits easier to file, but not easier to win
The reporting suggests AI is reducing one barrier to entry: drafting. It is not reducing the full burden of litigation. Levy told Technology Review that bringing a lawsuit is a “complex, multifaceted task,” and drafting text is only one component. Evidence, timing, jurisdiction, service, settlement posture, and courtroom strategy still decide outcomes.
This is consistent with broader court experience. Judge Braswell said she can often understand AI-assisted arguments better than filings written without such help. Yet the same reporting found that self-represented litigants still lose far more often than represented parties, and AI has not changed that pattern.
One reason is that language models are good at producing plausible form, not reliable legal judgment. In legal services and government workflows, that creates a familiar risk profile: improved throughput on the front end, more review burden on the back end. It is similar to what many enterprises see when generative systems draft policy memos, claims summaries, or procurement responses before a human checks them.
From the Encorp playbook: In high-stakes workflows, the first governance mistake is treating polished output as validated output. Organizations using AI for legal or quasi-legal drafting need usage rules, review thresholds, and escalation paths before staff rely on generated text externally. A good starting point is a fractional AI leadership and strategy model that sets those controls early.
The Reddit example in the article makes the point vividly. A December 2024 post reportedly advised immigration applicants to use Microsoft Copilot to draft a writ of mandamus, pay a lawyer $150 to clean it up, and file in Vermont. The result was a surge from roughly 45 such self-filed cases a year before 2022 to more than 1,100 in 2024. That is not merely a user adoption story. It is a workflow redesign story, driven by cheap drafting assistance and low-friction distribution through online communities.
The privilege fight is now as important as the drafting question
The more significant legal issue may not be whether chatbots can draft a complaint. It may be whether conversations with them are protected at all. Judge William Garfinkel in Connecticut raised the possibility that chatbot interactions used to prepare a case may deserve some protection analogous to legal work product or privilege.
Courts are already split. As Technology Review reports, a federal court in Michigan held in February that a self-represented litigant’s conversations with ChatGPT were protected work product. The same day, a federal court in New York reached the opposite conclusion for documents generated with Claude, reasoning that Claude is not a lawyer and that users may lack a reasonable expectation of confidentiality.
That split tracks a larger issue in AI data privacy governance and enterprise policy design. If users paste facts, claims, draft arguments, or settlement positions into a public model, they may believe they are preparing legal work when they are actually disclosing sensitive information to a third-party system. Judge Braswell’s March ruling, which suggested chatbot use should remain off limits in discovery despite data collection concerns, shows that courts are still feeling out where privacy expectations begin and end.
For legal teams, compliance officers, and public-sector administrators, this is where AI risk management becomes concrete. The question is no longer whether staff use chatbots. The question is whether the organization has rules for what data can enter those systems, what must remain inside approved tools, and how generated material should be retained or audited.
Liability for bad AI legal advice is moving from theory to litigation
The next front is responsibility. Judges are starting to ask whether a chatbot dispensing legal guidance should bear something like a duty of care, even if it is not a lawyer. Judge Allison Goddard’s example in California is telling: a plaintiff in a slip-and-fall matter reportedly demanded $700,000 based on ChatGPT’s guidance, only to be corrected in court.
That kind of incident does not prove that AI legal advice is uniquely dangerous; bad advice has always circulated through forums, templates, and informal networks. But it does show how systems can produce confident, well-phrased errors at scale. In practice, that means mistakes may reach the court faster and with greater persuasive polish.
The pending lawsuit from Nippon Life Insurance Company against OpenAI pushes the issue further by alleging that ChatGPT effectively practiced law without a license when it helped reopen a settled dispute. OpenAI’s response, also cited in the report, is that ChatGPT is not a person and does not practice law. That leaves courts with an unresolved category problem: these tools are not attorneys, but they are increasingly performing attorney-adjacent tasks.
Lawmakers are reacting unevenly. New York introduced a bill in March that would bar chatbots from impersonating lawyers even with disclosure, and members of Congress have proposed broader restrictions on chatbots posing as licensed professionals. Those proposals have not yet set a stable national rule, but the direction of travel is clear: AI support agents that touch regulated advice are moving into a tighter accountability environment.
What courts and organizations should watch next
The practical issue now is governance, not novelty. Courts need clearer standards for AI-assisted filings, privilege claims, and sanctions when generated content contains fabricated authority. Organizations adopting AI implementation services or custom AI integrations for legal, claims, or policy work need human review rules, staff training, and explicit limits on what external models may see.
The next phase of this story will likely be shaped less by better drafting models than by case law, courtroom procedure, and internal controls. AI legal advice is making access to filing easier; whether it makes justice better will depend on how institutions set boundaries around its use.
Martin Kuvandzhiev
CEO and Founder of Encorp.io with expertise in AI and business transformation