Ninth Circuit Warns Against AI Hallucinations in Filings
In LNU v. Blanche (No. 24-4790), a Ninth Circuit panel of Judges Richard Paez, Carlos Bea, and Danielle Forrest issued a disciplinary order against attorneys Mike Singh Sethi and William Rounds after their briefs cited nonexistent cases, misattributed quotations, and misrepresented real authorities. According to the opinion and reporting by Bloomberg Law, Law360, and Reason, the court suspended both attorneys from practice before the Ninth Circuit for six months, fined them $2,500 each, and required them to notify clients, opposing counsel, and presiding judges in their other cases. The attorneys initially blamed typographical errors before conceding the fabricated citations likely came from an unlicensed law-school graduate at the firm who used AI without authorization. The opinion distinguishes AI "fabrications" from "inaccuracies" and tells lawyers to read everything they cite and disclose hallucinations promptly.
What happened
In LNU v. Blanche (No. 24-4790), a Ninth Circuit panel of Judges Richard Paez, Carlos Bea, and Danielle Forrest issued a disciplinary order against attorneys Mike Singh Sethi and William Rounds, whose briefs contained "nonexistent cases, misattributed quotations, and gross misrepresentations of real cases," according to the opinion and reporting by Reason, Bloomberg Law, and Law360.
The sanctions
According to Law360 and the court opinion, the panel suspended both attorneys from practice before the Ninth Circuit for six months, imposed fines of $2,500 each, and ordered them to send the order to clients, opposing counsel, and presiding judges in their other matters. Reason reports the briefs cited cases that do not exist, including a fabricated "Eduardo v. Garland" and "Lay v. Holder," and attributed quotations to real opinions where the language does not appear. The attorneys first described the problems as typographical errors before conceding the citations likely came from an unlicensed law-school graduate at the firm who had used generative AI without authorization.
What the court said about AI
The opinion identifies two categories of generative AI error: "fabrications," where a tool invents cases or quotations, and "inaccuracies," where it cites real authorities but misstates them. The court warns that inaccuracies can be subtler and more dangerous because they survive cursory review, and instructs attorneys to be aware of the risks of overreliance on generative AI, read everything cited in a filing, and disclose hallucinations quickly and transparently.
Editorial analysis
Public discipline tied directly to AI hallucinations raises the operational and professional stakes for lawyers using generative tools. The court's emphasis on hard-to-detect inaccuracies points to growing demand for citation-grounding and automated verification in legal workflows, and to likely follow-on guidance from other courts and bar associations.
Key Points
- 1The Ninth Circuit suspended two attorneys from practice before the court for six months and fined them $2,500 each over briefs containing fabricated and misattributed authorities, per the opinion and Law360.
- 2The opinion distinguishes AI "fabrications" (invented cases or quotes) from "inaccuracies" (real authorities cited incorrectly), warning the latter are subtler and harder to detect.
- 3Editorial analysis (generic industry): public discipline tied to AI hallucinations raises the operational stakes for legal AI use and increases demand for citation-verification tooling and disclosure practices.
Scoring Rationale
A federal appellate court imposing concrete sanctions, including six-month suspensions, over AI-fabricated citations is a notable legal precedent with direct consequences for practitioners and clear implications for legal-AI verification tooling. Coverage across major legal outlets reinforces its significance, though its technical impact on core ML research is indirect.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
