Lacey v. State Farm Gen. Ins. Co.

The timeline

• 2024An insurance coverage suit proceeds in the Central District of California, with Ellis George LLP and K&L Gates LLP involved for the plaintiff side, as reported.
• Early 2025Attorneys at Ellis George use CoCounsel, Westlaw Precision and Google Gemini for legal research; K&L Gates, as cooperating counsel, files the resulting brief without independently checking the citations, as reported.
• May 2025The court orders $31,100 in fees and costs, joint and several against the two firms, after finding roughly a third of the brief's citations, nine of 27, were nonexistent or misstated the law. The Westlaw citation for the order is dated 5 May 2025; reporting describes the order of 6 May 2025.
• 23 Jul 2025Judge Manasco cites the $31,100 figure in Johnson v. Dunn, 792 F. Supp. 3d 1241, 1266 (N.D. Ala. 2025), independently confirming the amount.

What the AI did

As reported, the tools involved were not consumer chatbots but products sold to law firms for legal research. They still produced citations to cases that did not exist and statements of law the cited cases did not contain. The reporting also records that one of the tools was used to analyze the brief itself for “obviously bad caselaw”, and flagged nothing.

What the court did, and why

The court ordered $31,100 in fees and costs, joint and several, against both firms. Source for that award: Johnson v. Dunn, 792 F. Supp. 3d 1241, 1266 (N.D. Ala. 2025), citing Lacey, 2025 WL 1363069, at *5. The order text itself is UNVERIFIED, not read at source, so nothing here is quoted from it. The court’s own words below, and the finding on the citation count, reach this page through contemporaneous reporting rather than the document, and are UNVERIFIED on that basis.

“contained numerous false, inaccurate, and misleading legal citations and quotations”, as quoted in contemporaneous reporting

What it changed

Lacey is the row in this record most often cited inside large firms, because the defense that failed was the expensive one: a recognized legal-research vendor, plus a second firm that trusted the first. Nine wrong citations out of 27 is a one-in-three failure rate in work product that two firms touched. The lesson later cases keep repeating, grounded tools lower the error rate but do not transfer the duty to read the cited opinion, is the one this order priced at $31,100.

Sources

• Contemporaneous reporting on the order, the firms and the tools (Clearbrief), as reported
• Johnson v. Dunn, 792 F. Supp. 3d 1241 (N.D. Ala. 2025) (PDF), citing Lacey, 2025 WL 1363069, at *5, for the $31,100 award

Related

• The full sanctions record
• Wadsworth v. Walmart, the same lesson with a firm's own tool
• Couvrette v. Wisnovsky, where the fee-award curve ends
• Every verified AI standing order

Checked 4 August 2026. The $31,100 award and its joint-and-several structure are confirmed at a pin cite in a published federal opinion. The order text was not read at source, so the citation count, the named tools, the named firms and the quoted passage are UNVERIFIED and rest on the linked coverage. If the order is recovered, UNVERIFIED is replaced with the exact source and container. Information, not legal advice.