Skip to content

analysisAI HallucinationsAI in Courts

Ellis George and K&L Gates: the $31,100 AI-research sanction, and the re-filing that made it worse

A C.D. Cal. Special Master sanctioned Ellis George and K&L Gates $31,100 after an AI-drafted brief cited fake cases and was re-filed without disclosure.

Edited and verified by Cognesio LLP

Researched with AI assistance · sources verified by Cognesio LLP · How this was made ↓

In the US District Court for the Central District of California, a Special Master ordered the law firms Ellis George LLP and K&L Gates LLP to pay $31,100 after a discovery brief built on undisclosed AI research cited cases that did not exist. The order in Lacey v. State Farm General Insurance Co., No. 2:24-cv-05205-FMO-MAA, was signed by Special Master Michael R. Wilner, a retired US Magistrate Judge, on 5 May 2025 and filed on 6 May 2025. It is one of the few rulings on the tracker where the tools named include paid legal-research products, not just a general chatbot, and one of a small group, with Mata v Avianca and Couvrette v. Wisnovsky, where the sanction turns on what the lawyers did after the problem was flagged.

What happened?

A partner at Ellis George used AI tools to produce a research outline for a ten-page supplemental brief on privilege. Lawyers at K&L Gates wrote the outline into the brief without checking it. About nine of the 27 citations were incorrect in some way, from wrong pin cites to invented cases; at least two authorities did not exist, and several quotations were not in the opinions they were attributed to.

The Special Master’s account is short and specific. Trent Copeland of Ellis George “used various AI tools to generate an ‘outline’ for the supplemental brief.” His declaration, attached to the order, names them: CoCounsel, Westlaw’s AI tool, and Google’s Gemini. He sent the outline to K&L Gates, whose lawyers “incorporated the material into the brief” and, on the sworn evidence, did not know AI had been used and did not ask. “No attorney or staff member at either firm apparently cite-checked or otherwise reviewed that research before filing the brief with the Special Master.”

Then came what the order treats as a further wrinkle. The Special Master could not confirm two of the cited authorities and emailed the lawyers. K&L Gates re-submitted the brief the same day with those two citations removed “but with the remaining AI-generated problems in the body of the text”. An associate’s covering email said the two errors had been “inadvertently included”. Nothing in it mentioned AI. The Special Master found out how the brief had been written only after issuing a formal order to show cause.

What did the Special Master decide?

He found that the lawyers “collectively acted in a manner that was tantamount to bad faith”, struck the plaintiff’s supplemental briefs, refused all of the discovery relief they sought on the privilege issue, and ordered the two firms jointly and severally to pay $31,100 to the defence. He declined to sanction any individual lawyer.

The standard was the Ninth Circuit’s test in Fink v. Gomez, 239 F.3d 989 (9th Cir. 2001), under which recklessness combined with an improper purpose can be treated as bad faith. The order works through three failures: the undisclosed use of AI to write the first draft, the failure to cite-check it, and the re-submission of a still-defective brief “without adequate disclosure of the use of AI”. The third is the one the Special Master calls “perhaps most egregiously” wrong. On the first he wrote:

“Even with recent advances, no reasonably competent attorney should out-source research and writing to this technology – particularly without any attempt to verify the accuracy of that material.”

The order’s closing paragraph explains why the sanction was more than a warning:

“Directly put, Plaintiff’s use of AI affirmatively misled me. I read their brief, was persuaded (or at least intrigued) by the authorities that they cited, and looked up the decisions to learn more about them – only to find that they didn’t exist. That’s scary.”

Where does the $31,100 come from?

The figure is two awards added together. The larger part, about $26,100, is the Special Master’s own fees for dealing with the defective briefs, which the district court’s appointment order had allowed him to shift. The defence had advanced those fees to JAMS, so the firms were ordered to reimburse the defence within 30 days. The smaller part is $5,000 towards the defence’s costs of the briefing, cut down from a claimed figure that “approached $25,000” because the Special Master did not think full compensation was needed for deterrence.

Two things the order does not do matter as much as what it does. It does not impose any penalty on the individual lawyers, whose admissions the Special Master called “full, fair, and sincere”. And it does not refer anyone to the State Bar of California. The money falls on the firms, not the client: the order records that Mrs Lacey “is clearly not at fault for the AI debacle” and “will not, however, be financially responsible for the monetary awards”.

Why it matters

Most sanctions rulings on the tracker turn on a single bad filing. This one turns on the second. The Special Master had flagged a problem, the firms had the chance to check the whole brief and say how it had been written, and instead they removed the two citations he had noticed and described the rest as inadvertent. The order draws the contrast with United States v. Cohen, 724 F. Supp. 3d 251 (S.D.N.Y. 2024), where Judge Furman declined to find bad faith on the part of the lawyer who filed citations invented by Google Bard: he learned of the problem only from the court’s order to show cause and had never been given the chance to withdraw them. What the Special Master punished here was not the first mistake but the chance to fix it that went unused.

The tools named are also a data point. This was not a lawyer pasting a ChatGPT answer into a brief. The outline came from CoCounsel and Westlaw Precision, products listed in the legal-research category of the SafeLegalAI tools directory rather than general-purpose chatbots, plus internet research in Google’s Gemini. The order does not say which tool produced which error, and the tracker does not guess. What it records is that paid legal-research tools were in the chain and the verification step was skipped anyway.

Sources