A federal judge in Kansas has handed down $12,000 in penalties after discovering that court submissions in a patent dispute were padded with quotations and legal citations that simply did not exist — inventions traced back to artificial intelligence.
The sanctions stem from filings made on behalf of Lexos Media IP in its patent infringement case against online retailer Overstock.com. The documents included fabricated case law and misquoted material, which the court found had been generated by AI and then filed without proper verification.
While only one attorney admitted to using AI to prepare the filings, the court ruled that responsibility did not stop there. Every lawyer who signed the documents was faulted for failing to independently review and confirm their accuracy.
In a sharply worded order, the judge stressed that the dangers of relying on unverified generative AI for legal research are no longer obscure or theoretical. Given the growing number of cases involving bogus AI-generated citations, lawyers are expected to know better — and to act accordingly.
The court described the recent surge of disciplinary actions tied to AI misuse as alarming, noting that fictional legal authorities produced by such tools have repeatedly found their way into official filings across the country.
The issue came to a head after the judge ordered five attorneys involved in the case to explain defects in multiple submissions, including nonexistent quotations, phantom citations, and misleading representations.
One attorney was fined $5,000 after acknowledging that he had used an AI chatbot without checking its output, citing personal pressures at the time. In addition to the fine, he was instructed to share the ruling with state disciplinary authorities and to formally outline steps his firm will take to prevent a repeat episode.
Calling the experience “an embarrassing lesson,” the lawyer said firms should not deploy AI in legal work without strict internal controls and review policies.
Two other attorneys were fined $3,000 each for signing off on filings they failed to scrutinize, while local counsel was ordered to pay $1,000 for neglecting to verify the citations.
The ruling adds to a growing body of courtroom warnings: AI may be fast and fluent, but unchecked, it can also be confidently wrong — and courts are making it clear that the consequences will land squarely on the humans who submit the work.
