Background
Two Orange County immigration attorneys, Mike Singh Sethi and William Rounds, filed appellate briefs in the Ninth Circuit that cited two cases that do not exist — Eduardo v. Garland and Lay v. Holder — and misquoted language from several real cases. The briefs were drafted by unlicensed law school graduates working at the firm, and no licensed attorney verified the citations before the briefs were signed and filed.
When the fabrications surfaced, Sethi initially characterized them as “typographical errors” and offered replacement citations. At oral argument, Rounds was asked directly whether generative AI had been used to draft the brief. He denied it three times, then conceded it was “possible,” and ultimately disclosed that the unlicensed drafter had produced the brief without any attorney having checked the cited authorities.
The Court’s Holding
The panel sanctioned both attorneys. The most consequential line in the opinion is that “the rules are violated at the point of signing and filing,” not at the point of research. An attorney’s signature on a brief is a personal attestation that every cited authority has been read and confirmed — a duty the court treated as non-delegable, whether the underlying draft came from a junior associate, a contract paralegal, or a large language model.
The court found violations of Federal Rule of Appellate Procedure 28 (which governs the contents of a brief) and California Rules of Professional Conduct 1.1 (competence), 1.3 (diligence), 3.1(a)(2) (meritorious claims), and 3.3(a)(1) (candor to the tribunal). The panel was careful to note that using generative AI is not itself unethical. The violation lies in letting unverified AI output reach the court under an attorney’s signature.
The court also singled out the courtroom denials. The opinion treats the repeated misrepresentations — first calling fabrications “typos,” then denying AI involvement three times under direct questioning — as a candor violation distinct from, and more serious than, the underlying citation errors.
Sanctions Imposed
- $2,500 monetary fine on each attorney
- Six-month suspension from practice before the Ninth Circuit
- Mandatory AI-use disclosure statement on every brief filed in the Ninth Circuit for the next two years
- Referral to the State Bar of California for further discipline
Key Takeaways
- The signature is the trigger. The ethical duty attaches when a brief is signed and filed, regardless of who or what drafted the underlying text. Citation verification cannot be delegated.
- AI use is not the violation; unverified output is. The court explicitly rejected the idea that using generative AI for drafting or research is itself sanctionable. What gets attorneys sanctioned is failing to check the work.
- Candor failures compound. The panel treated the repeated denials at oral argument as a separate, aggravating violation. Honest disclosure when fabrications surface is materially less risky than denial.
- Disclosure orders are now a real remedy. Forced AI-use disclosure on future filings is becoming a standard add-on to monetary and suspension penalties.
Why It Matters
This is the Ninth Circuit’s most direct statement to date on generative AI in legal practice, and it sets a technology-neutral standard: an attorney is responsible for every citation in every brief that bears their signature. The decision is significant because it refuses to treat AI as either a special excuse (“the chatbot did it”) or a special crime (“AI use is unethical”). It treats AI exactly like any other drafting tool — useful, but no substitute for the attorney’s own verification.
For California firms, especially those with high-volume practices that rely on junior or unlicensed staff to draft, this decision raises the cost of skipping the citation-check step. A six-month suspension from a circuit, combined with a state bar referral, can end a litigation practice. Firms that build AI into their workflow should pair it with a mandatory, documented citation-verification step before any brief is signed. Solo practitioners and small firms should treat AI-drafted research the same way they would treat a memo from a brand-new associate — useful, but not yet trustworthy.