For more than two years, the legal profession’s reckoning with generative AI hallucinations played out almost entirely in the trial courts. District judges across the country struck briefs, ordered show-cause hearings, and imposed monetary sanctions on lawyers who filed motions citing cases that did not exist. We catalogued that wave in our Legal AI Reckoning analysis and examined how even elite firms were not immune when we covered the Sullivan & Cromwell hallucination filing. But those orders, however numerous and however pointed, shared a structural limitation: they were trial-court decisions. They bound the parties before them. They did not bind anyone else.

On June 3, 2026, that changed. In LNU v. Blanche, No. 24-4790, the United States Court of Appeals for the Ninth Circuit issued a published, precedential opinion sanctioning two attorneys for filing briefs riddled with AI-generated fabrications. This is the appellate precedent the earlier coverage was waiting for. A published circuit opinion is binding law throughout the nine western states and two Pacific territories within the Ninth Circuit’s jurisdiction, and persuasive authority everywhere else. It is the kind of decision that district judges cite, that bar disciplinary committees lean on, and that opposing counsel attach to their own sanctions motions. The hallucination problem now has appellate teeth.

Why a Published Appellate Opinion Changes the Calculus

The distinction between an unpublished trial-court order and a published circuit opinion is not academic. When a federal district judge sanctions a lawyer for citing Bourhill v. Young as if it governed an asylum claim, that order resolves a dispute. It may embarrass the lawyer, it may cost money, and it may be reported in the legal press. But it sets no rule. The next lawyer in the next courtroom can argue, with a straight face, that the prior order turned on its particular facts.

A published opinion of a circuit court of appeals does something different. It announces a legal standard that lower courts within the circuit must follow. The Ninth Circuit panel — Judges Richard Paez, Carlos Bea, and Danielle Forrest — wrote LNU v. Blanche for publication precisely because the court wanted to establish, as a matter of governing law, when and how an attorney’s use of generative AI crosses the line into a sanctionable breach of professional duty. The opinion is no longer one judge’s reaction to one bad brief. It is the framework.

That framework matters far beyond immigration practice, far beyond California, and far beyond litigation. Any regulated organization that uses generative AI to produce documents on which a professional attests — pleadings, regulatory filings, audit representations, sworn certifications — should read this opinion as a statement of where the duty attaches and what disclosure obligations follow.

The Facts: From Hallucinated Citations to a Failure of Candor

The underlying matter was an immigration appeal. The petitioner, identified in the caption only by the placeholder “LNU” (the litigation convention for “Last Name Unknown”), was represented before the Ninth Circuit by Mike Singh Sethi of the Sethi Law Group in Orange, California, and William Rounds of Bill Rounds Attorney at Law PC, also in Orange, California.

The briefs they filed contained what the panel described as nonexistent cases, misattributed quotations, and gross misrepresentations of real authority. In other words, the classic signature of a generative AI model asked to find supporting law: confident, fluent, properly formatted citations to decisions that were never decided and quotations that were never written.

When the problem surfaced, the attorneys filed a Motion to Correct. Critically, they did not at that point disclose that the fabrications were the product of generative AI. Instead, they characterized the nonexistent cases as the result of typographical or copy-paste errors. That characterization is what transformed a recoverable mistake into a sanctionable one. The court ordered Sethi and Rounds to show cause why they should not be sanctioned, suspended, or disbarred from practice before the court. After considering their response, the panel imposed discipline.

The sanctions, entered under Federal Rule of Appellate Procedure 46(b) and Ninth Circuit Rule 46-2, were as follows:

  • Sethi and Rounds were each personally sanctioned $2,500, payable within twenty-one days to the Clerk of the US Courts.
  • Each was suspended from practice before the Ninth Circuit for six months.
  • Both were ordered to provide a copy of the order to their clients, opposing counsel, and the presiding judge in all of their other pending cases — a transparency obligation that ensures the discipline follows them into every matter they are currently handling.
  • The entire firm was ordered to include, in all future filings, a statement under penalty of perjury addressing whether generative AI was used, disclosing the name of the tool used, and certifying that the signing attorney personally reviewed the filing.

The panel also referred the matter to the State Bar of California, where independent disciplinary consequences may follow.

The Holding: The Violation Attaches “At the Point of Signing and Filing”

The single most important sentence in the opinion is not about money or suspension. It is about timing. The court drew a clean line that every lawyer and every legal-adjacent professional should commit to memory:

The rules are not violated at the point of research and drafting, but at the point of signing and filing.

This is the holding that gives the opinion its lasting significance. The Ninth Circuit went out of its way to clarify what it was not sanctioning. The panel expressly stated that it was not imposing discipline for the simple fact that Sethi, Rounds, or their subordinates used generative AI. Using an AI tool to research or draft is not, by itself, a violation of any rule. The court refused to treat generative AI as contraband.

What the court sanctioned was the filing of false authorities and, even more sharply, the failure of candor that followed. An attorney’s signature on a brief is an attestation — under Rule 11’s federal-court analogue and the Federal Rules of Appellate Procedure’s candor obligations — that the signer has personally reviewed the contents, including the cited authorities, and that they are accurate. The provenance of the error is, at the moment of signing, beside the point. As the panel put it, if an attorney files a brief with cases or quotations that do not exist, or that completely misrepresent what a real authority stands for, it generally does not matter how the false material got there. The duty was to verify before signing. Signing without verifying is the breach.

This framing is doctrinally elegant and practically devastating. It means a lawyer cannot defend a hallucinated citation by saying “the AI produced it.” The AI’s role is upstream of the duty. The duty crystallizes at signature.

The Duty to Correct — and to Disclose the Source

The opinion adds a second, equally consequential holding about what happens after a mistake is discovered. When an attorney learns of any error in a filing — including a generative AI hallucination — the panel held that the attorney should immediately alert the court and opposing counsel of the error and disclose its source. Disclosure of the source is not optional politeness. In circumstances like these, the court explained, the duty of candor requires a transparent disclosure of the source of the errors.

This is where Sethi and Rounds compounded their exposure. The panel was explicit that the six-month suspension, as opposed to a lesser sanction, was owed to a repeated failure of candor — the decision to recast AI fabrications as typos rather than confess their origin. The opinion all but tells future litigants that there is no upside to concealment: had the attorneys disclosed in their Motion to Correct that AI had been used against firm policy and apologized, the outcome would likely have been materially different. The cover-up, as is so often the case, drew a harsher penalty than the underlying conduct.

The Forward-Looking Mandate: Perjury-Backed AI-Use Disclosure

The most operationally significant piece of the order is prospective. The Ninth Circuit did not merely punish past conduct; it imposed a continuing compliance regime on the entire firm. Going forward, every filing by Sethi, Rounds, or any attorney at the firm must contain a statement, made under penalty of perjury, that:

  1. addresses whether generative AI was used in preparing the filing;
  2. discloses the name of the tool used; and
  3. certifies that the signing attorney has personally reviewed the filing.

Three features of this mandate deserve emphasis. First, it is backed by penalty of perjury, which elevates a false certification from a Rule 11 problem to a potential criminal exposure. Second, it requires affirmative naming of the specific tool — not a vague acknowledgment that “AI may have been used,” but identification of the product. Third, it ties the certification to personal review by the signing attorney, foreclosing the delegation-and-trust model in which a partner signs what an associate or a contract attorney assembled without independently confirming the authorities.

While this particular order binds only one firm, its content is a template. Courts and bar regulators now have a published appellate decision endorsing perjury-backed AI-use certification as an appropriate remedy. Standing orders requiring exactly this kind of disclosure have been proliferating at the trial-court level; LNU v. Blanche gives that practice an appellate imprimatur.

And It Is Not Just the Appellate Courts

The Ninth Circuit’s opinion did not land in isolation. Five days later, on June 8, 2026, U.S. District Judge Sharion Aycock of the Northern District of Mississippi took an even more dramatic step in a contractual dispute between Tom Withers III and the City of Aberdeen. Confronted with fabricated legal citations in filings from both sides, Judge Aycock removed all four lawyers from the case.

The four attorneys were Kathleen Wilson, a Louisiana-based lawyer admitted pro hac vice; Shauncey Hunter Ridgeway of Jackson, Mississippi; Mark McClinton of New Albany, Mississippi; and Kathryn Young Williams, a Texas-based lawyer admitted pro hac vice. The judge found that all four had violated Federal Rule of Civil Procedure 11 by failing to verify their legal authorities before filing — a textbook articulation of the same duty the Ninth Circuit located at the point of signing. Reporting on the matter described the court’s concern that the lawyers had been “blindly relying on technology,” producing “hallucinatory citations.” Beyond removal from the case, the two out-of-state practitioners were barred from appearing before the Northern District of Mississippi for two years.

The Mississippi order is not precedent in the way the Ninth Circuit opinion is. But it is a data point that makes the trend unmistakable: appellate and trial courts, in different circuits, within a single week, treated AI-hallucinated citations as serious professional misconduct warranting removal, suspension, and monetary penalties. The judiciary has converged on a consistent message, and lawyers can no longer plausibly claim the rules are unsettled.

The practical lessons of LNU v. Blanche extend to any law firm, in-house legal department, or regulated enterprise that uses generative AI to draft documents on which a professional signs or certifies. The opinion does not require abandoning AI. It requires building the controls that the duty of candor has always implied but that the speed and fluency of AI now make urgent.

Verify every authority before signing. The non-negotiable takeaway is that citation verification is not a clerical task that can be skipped under deadline pressure. Every cited case must be pulled, read, and confirmed to stand for the proposition asserted — by a human, before the document is filed. The signature is the attestation. Treat it that way.

Adopt and enforce an AI-use policy — and follow it. The Sethi firm had a policy against using AI for the work in question; the problem was that the policy was not honored, and then its violation was concealed. A policy that exists on paper but is not enforced is worse than no policy, because it establishes a standard the firm can be shown to have breached. Define where generative AI may and may not be used, require disclosure of its use internally, and audit compliance.

Build AI-use certifications into your filing workflow now. Do not wait for a standing order or a sanctions motion. Firms should consider adopting their own internal certification — a checkpoint requiring the signing attorney to confirm, before filing, whether AI was used, which tool, and that all citations were personally verified. This is precisely the regime the Ninth Circuit imposed as a sanction; voluntarily adopting it is cheap insurance.

Supervise subordinates and vendors. The opinion’s reference to “subordinates” is a pointed reminder that the supervising attorney owns the work product. The duty cannot be delegated to a junior associate, a paralegal, a contract attorney, or a software tool. Partners who sign must verify, or must ensure that someone they directly supervise has verified and can attest to it.

When you discover an error, disclose immediately — and disclose the source. This is the lesson the sanctioned attorneys learned the hard way. The moment a hallucinated citation surfaces, the correct response is prompt notice to the court and opposing counsel, accompanied by candid disclosure that the error originated in generative AI. Reframing a fabrication as a typo is the conduct that converted a $2,500 problem into a six-month suspension and a bar referral.

Extend the discipline beyond litigation. Although LNU v. Blanche arises from court filings, the underlying principle — that liability attaches when a professional attests to content they did not verify — maps directly onto regulatory submissions, audit certifications, sworn affidavits, securities disclosures, and any other context where AI-drafted text carries a human signature backed by professional or legal accountability. The point of signing is the point of exposure everywhere, not only in the Ninth Circuit.

Conclusion

LNU v. Blanche is a short opinion with a long reach. It did not outlaw generative AI in legal practice, and it explicitly declined to treat the use of AI as the wrong. What it did was locate, with precision and as binding precedent, the moment at which responsibility attaches: not when the model drafts, but when the lawyer signs and files. It paired that holding with a continuing duty to correct and to disclose the source of any error, and it imposed a perjury-backed AI-use certification that courts and regulators across the country now have appellate authority to replicate.

For two years the profession could tell itself that the hallucination cases were isolated, fact-bound, and confined to the careless few. After June 3, 2026 — and after Judge Aycock’s order five days later — that story no longer holds. The standard is published. The duty attaches at signature. And the only safe assumption for any professional using generative AI is that a court will hold them personally accountable for every word over which they sign their name.

This article is provided for informational purposes only and does not constitute legal advice.