Roanoke lawyer's AI-generated legal research derailed assault case dismissal
A federal judge dismissed a 133-page lawsuit after finding that the Roanoke attorney who filed it relied on artificial intelligence to fabricate case law and misquote legal precedent. U.S. District Judge Michael Urbanski ruled the filing contained "hallucinated law ostensibly obtained from generative AI tools" and a "concerning number of false quotations, nonexistent cases and misstatements of law."
The case involved Raymond Lewis Sherwood III, a former Botetourt County school bus driver who pleaded no contest to two counts of misdemeanor assault and battery after being charged with 25 counts of assaulting children. His attorney, Jon Clark, sued the prosecutor, sheriff, and school board alleging malicious prosecution, abuse of power, and defamation.
Urbanski dismissed all 20 counts of the lawsuit on the merits. But he flagged Clark's AI use as the source of fundamental legal errors that undermined the entire case. Clark incorrectly argued that police were required to file a sworn affidavit to support misdemeanor charges - they are not. He also claimed Sherwood could challenge the truthfulness of allegations after pleading no contest, which the law does not permit.
Clark acknowledged the mistakes in an email, writing: "The responsibility for verifying the accuracy of legal citations ultimately rests with me, and I take that responsibility seriously." He said the experience "has reinforced the importance of careful verification and review when using evolving legal research technologies."
A growing problem across courts
The Roanoke case is believed to be the first documented instance of AI-generated legal errors in the city's federal court. But attorneys nationwide are increasingly running into trouble when they use AI for legal research.
Brett Marston, managing partner of Gentry Locke and president of the Virginia State Bar, said the problem is "surprisingly frequent." He said the state bar has been examining what is referred to as AI hallucination - when generative AI tools produce false information presented as fact.
A global database compiled by French legal researcher Damien Charlotin called AI Hallucination Cases tracks slightly more than 1,000 cases in the United States. Twelve are from the Western District of Virginia. Most involve pro se litigants - people representing themselves - rather than licensed attorneys.
Jessiah Hulle, a Roanoke attorney who follows AI misuse in legal practice, said accurate data is hard to come by. "It's often by word of mouth when you find out this is happening," he said.
Consequences for attorneys
When lawyers use AI to make spurious arguments or cite fabricated cases, judges can impose sanctions, hold them in contempt of court, or refer them to the state bar for discipline. "It's something that the judges will take very seriously," Marston said.
A federal judge in North Carolina recently reprimanded an assistant U.S. attorney for filing an AI-generated brief riddled with errors. The prosecutor announced his resignation at the hearing.
In a whistleblower case in Harrisonburg, U.S. District Judge Thomas Cullen found that an attorney's filing contained "citations to sources and quotations that appear not to exist." The opposing party called it an example of "ChatGPT run amok." Cullen threatened sanctions but ultimately did not impose them after the attorney said it was an accident involving miscited real cases.
Clark has not been sanctioned in the Sherwood case.
Why attorneys turn to AI
Many lawyers are drawn to AI for its speed in generating results. But that speed creates a trap, said Les Bowers, a Charlottesville attorney on the Virginia Trial Lawyers Association's executive committee.
The risk rises when practitioners "fail to be lawyers" by not double-checking information before inserting it into court documents. "The main takeaway is that no law firm or lawyer is immune from this being a problem," Bowers said.
Judges have shown more flexibility with pro se litigants. U.S. District Judge Jasmine Yoon, writing in a case where a man challenging a speeding camera ticket used what appeared to be fabricated citations, acknowledged "the hurdles that pro se litigants face when pursuing claims in federal court." Still, she wrote, "submitting a filing containing false citations - whether created by generative artificial intelligence or not - is unacceptable."
For legal professionals navigating these tools, resources on AI for Legal work and the AI Learning Path for Paralegals address practical applications and verification standards in legal document review and research automation.
Your membership also unlocks:
