A court filing in a civil dispute against State Farm has become another warning about using AI for legal research without verification. The problem was not only that citations were wrong. Judge Michael Wilner said the flawed material came close to shaping a court order.
The sanction was aimed at law firms representing former Los Angeles County District Attorney Jackie Lacey. Wilner, a retired US magistrate judge serving as special master in US District Court for the Central District of California, ordered penalties totaling $31,100 after finding that a supplemental brief contained fake AI citations, inaccurate legal references, and phony quotations.
How the fake citations reached the court
Lacey is represented by K&L Gates and Ellis George LLP. Wilner described the group as “a large team of attorneys” and called the episode “a collective debacle.” The filings were part of a discovery fight over State Farm’s assertion of various privileges.
The source of the problem was an AI-generated “outline.” According to Wilner, the lawyers admitted that Mr. [Trent] Copeland, an attorney at Ellis George, used various AI tools to create that outline for the supplemental brief. The document included the faulty legal research, and lawyers at K&L Gates later incorporated the material into the brief.
Wilner found that no attorney or staff member at either firm apparently cite-checked the research before the brief was filed with the special master. He also wrote that the K&L Gates lawyers did not know Copeland had used AI to prepare the outline, and did not ask him.
The resulting brief had a significant error rate. Wilner found that approximately nine of the 27 legal citations in the ten-page brief were incorrect in some way. At least two of the cited authorities did not exist at all. Several quotations attributed to judicial opinions were also phony or did not accurately represent the cited materials.
The judge says he was nearly misled
The unusual part of the order is how plainly Wilner described his own reaction. He said he initially found the cited authorities persuasive or at least interesting enough to investigate. When he searched for the decisions, he discovered that they did not exist.
“Directly put, Plaintiff’s use of AI affirmatively misled me,”
That statement matters because the fake citations were not caught only by an opposing party after the fact. Wilner himself noticed that some citations could not be verified. Even so, he wrote that the bogus materials almost found their way into a judicial order.
Wilner also rejected the idea that the error was harmless because the AI output was close to the law in substance. He wrote that the “AI hallucinations weren’t too far off the mark in their recitations of the substantive law,” but he treated that as a weak defense. In his view, legal material that looks plausible can still be dangerous when it is not real.
A second filing still carried AI errors
The problem did not end with the first warning. During his initial review, Wilner could not confirm two citations and emailed the lawyers about them. K&L Gates then resubmitted the brief without those two citations.
But the revised brief still contained other AI-generated problems. Wilner later issued an OSC, or order to show cause, seeking a fuller explanation. Only then did he learn that AI had been used and that the second version still contained more made-up citations and quotations beyond the two initial issues.
The lawyers later submitted sworn statements and the AI-generated outline itself. Wilner wrote that those materials clarified the chain of events that led to the false filings. He also noted that the declarations included “profuse apologies and honest admissions of fault.”
Legal scholar Eugene Volokh wrote about the incident and noted that both firms involved had strong reputations in his view. He described one as a massive 1,700-lawyer national firm and the other as a smaller 45-lawyer predominantly California firm, and said the error was not characteristic of their work.
Why the sanctions hit the case and the firms
Wilner concluded that the lawyers collectively acted in a way “tantamount to bad faith.” He criticized Copeland’s undisclosed use of AI products and said no reasonably competent attorney should outsource research and writing to the technology without verifying the accuracy of the material.
He also faulted the K&L Gates lawyers for failing to validate the research they received. After he raised concerns about part of the brief, their response was to remove the two flagged citations and file a revised version that still contained a half-dozen AI errors. Wilner wrote that the conduct showed recklessness with the improper purpose of trying to influence his analysis of the disputed privilege issues.
The sanctions had practical consequences for the plaintiff’s position. Wilner struck the supplemental briefs on the privilege issue and declined to consider them. He also declined to grant the discovery relief the plaintiff had sought in those proceedings, including augmenting a privilege log, ordering production of materials, or requiring in camera review of items.
The financial penalty was directed at the firms, not the individual lawyers. Wilner ordered Ellis George and K&L Gates to pay $26,100 to the defense as reimbursement for fees paid to an arbitration and mediation firm, plus another $5,000 to cover some of the defense’s other costs.
He decided not to sanction or penalize the lawyers individually. Wilner wrote that their admissions of responsibility were full, fair, and sincere, and that he accepted their apologies. For future filings, however, the message of the order is direct: AI-generated legal research may appear convincing, but every citation and quotation still has to be checked before it reaches the court.