A federal court in Kansas has drawn a hard line on the careless use of artificial intelligence in legal filings, sanctioning a team of lawyers after discovering that their court submissions leaned on case law that simply did not exist.
The ruling came from U.S. District Judge Julie Robinson, who found that briefs filed on behalf of Lexos Media IP, LLC in a patent infringement dispute against Overstock.com were riddled with fabricated citations, invented quotations and distorted readings of real decisions. The problem, the court said, was not the use of AI itself—but the blind trust placed in it.
Two briefs filed in July 2025 triggered the fallout: one opposing a bid to exclude Lexos’ technical expert and another responding to a motion for summary judgment. Overstock’s reply flagged multiple “defective” citations, prompting closer scrutiny. What the court uncovered was troubling.
Among the errors were references to cases that could not be found in any federal database, including a supposed 2020 decision titled Hockett v. City of Topeka. Other citations attributed neatly packaged quotes to genuine Federal Circuit rulings—quotes that appeared nowhere in the actual opinions. Some decisions were cited for legal principles they never addressed at all.
Judge Robinson concluded that all five attorneys who signed the filings breached Rule 11 of the Federal Rules of Civil Procedure, which requires lawyers to confirm that their legal arguments are grounded in real law after a reasonable inquiry. That duty, the court stressed, applies equally to every name on the signature line.
One attorney, Sandeep Seth, acknowledged he was the primary drafter and admitted to using ChatGPT as a research shortcut while dealing with family medical issues. He described himself as inexperienced with generative AI and unaware that it could “hallucinate” authorities. The court was unmoved. Personal circumstances, Judge Robinson wrote, did not excuse outsourcing legal research to an unfamiliar AI tool without verification—especially when conventional research methods were readily available.
Sanctions ranged from $1,000 to $5,000 per lawyer, calibrated to each attorney’s role and level of responsibility, along with additional disciplinary measures.
The message from the bench was blunt: generative AI has a place in modern legal practice, but unchecked reliance on its output does not. Failing to verify AI-generated material, the court warned, is a serious professional lapse—one that wastes judicial time and unfairly burdens opposing parties.
In short, the court made clear that while machines may draft, humans remain accountable.
