top of page
background-3.2.jpg
Search

Lnu v. Blanche: The Duty Attaches at the Signature

Writer: maasdev
maasdev
2 hours ago
6 min read

The Ninth Circuit becomes the first federal appellate court to suspend lawyers for filing AI-fabricated authorities, and it fixes accountability at the moment counsel signs and files.


By Prof. Barry Appleton, FCIArb, TechCred Faculty Director, ABA TechCred Program, Co-Director and Distinguished Senior Fellow, New York Law School


The Ninth Circuit becomes the first federal appellate court to suspend lawyers for filing AI-fabricated authorities, and it fixes accountability at the moment counsel signs and files.




On June 3, 2026, the United States Court of Appeals for the Ninth Circuit drew the line that two years of trial-court orders had been circling. In Lnu v. Blanche, the court suspended two lawyers from practice before it, fined them, and, as a published precedent, explained at length exactly when a lawyer becomes answerable for a fabricated citation. The answer is not subtle. The duty attaches the moment the lawyer signs and files.


For two years, the profession could treat the hallucination problem as a scatter of mostly trial-court embarrassments. District judges struck briefs and imposed fines, but those orders bound no one beyond the courtroom that issued them. The appellate courts had begun to stir. The Second Circuit referred counsel to its grievance panel in Park v. Kim, and in January the New York Appellate Division imposed sanctions in Deutsche Bank National Trust Co. v. LeTennier, the first New York appellate decision to discipline fabricated AI authorities, where counsel filed twenty-three fabricated cases across five filings and kept filing them after being put on notice. Lnu raises the altitude again. It is the first federal court of appeals to suspend lawyers from practice for the conduct, and it does so in a published opinion that binds the nine western states and the Pacific territories and persuades elsewhere. The line that runs from Mata v. Avianca, Inc. through Park v. Kim and LeTennier now carries a suspension. The two appellate decisions converge on one aggravating fact: counsel who kept standing behind fabricated authority after the court put them on notice.


What happened


The underlying matter was an immigration appeal. The petitioners, identified in the caption by the docket convention “Lnu” for last name unknown, sought review of a Board of Immigration Appeals order. Their counsel, Mike Singh Sethi and William Rounds of Orange County, filed an opening brief that cited cases that do not exist.


Two of the citations were pure inventions. The brief relied on “Eduardo v. Garland” and “Lay v. Holder”, neither of which has ever existed. Two further citations pointed to real decisions, Kamalthas v. INS and Avendano-Hernandez v. Lynch, but attached to them were quotations that those opinions never contained.


Then came the conduct that decided the sanction. When the fabrications surfaced, Sethi filed a motion describing the two nonexistent cases as “typographical errors” and asked to swap in real decisions with similar names. The replacement cases did not support the propositions for which they had been cited. Sethi did not appear at oral argument. Rounds did, and when the panel asked whether generative AI had produced the errors, he denied it three times before conceding, minutes later and only under continued questioning, that it was possible. He then revealed that the briefs had been written by an unlicensed law-school graduate whose citations no admitted attorney had read.


The court imposed a six-month suspension from practice before it, a personal sanction of $2,500 on each attorney, and an order requiring both to send the decision to their clients, to opposing counsel, and to the presiding judge in every affected matter. It went further. For two years, every filing from the firm must certify, under penalty of perjury, whether AI was used and that a responsible attorney has personally verified each citation.


What the court actually held


Read the Lnu v. Blanche opinion for what it refuses to say before reading it for what it commands. The court did not hold that lawyers may not use AI. It described such use as “not inherently unethical or irresponsible.” Sethi and Rounds were not disciplined for using a machine. They were disciplined for what they did once the machine failed them.


Four holdings carry the decision.

  • The violation attaches at signing and filing. The court located the wrong not at the moment the tool drafts but at the moment the lawyer puts a name to the document and submits it. The signature is the representation. Everything that follows from a false citation flows from that act, regardless of who or what produced the words.

  • The duty of candor is continuing, and a quiet fix does not satisfy it. Substituting a real case for a hallucinated one does not cure the original false statement that the fabricated case existed. A lawyer who discovers a hallucination must disclose its nature and its source. Recasting a fabrication as a clerical slip is conduct unbecoming a member of the court’s bar.

  • AI errors and human errors are treated identically. A lawyer must read everything cited in a filing, whether a machine drafted it or not. The court pointed to a 2025 study finding that the leading commercial legal research tools still hallucinated answers between 17 and 33 percent of the time, and it described the resulting mistakes as the kind a first-semester law student might make but a licensed advocate may not.

  • The duty is personal and cannot be delegated. Responsibility cannot be pushed down onto unlicensed staff, nor can it be pushed sideways onto the tool. Asking an AI system to check its own work is not verification. The obligation runs to the lawyer whose name appears on the brief.


Why this reaches arbitration


Lnu is a litigation decision. It governs advocates in the federal courts of the western United States, and it persuades elsewhere. It is not an arbitration case. Read it beside ARIHQ anyway.


In April, in ARIHQ c. Santé Québec, the Quebec Superior Court annulled an arbitral award because the arbitrator used AI to generate reasoning that rested on hallucinated authorities. That decision placed personal, non-delegable accountability on the arbitrator through the personal-mandate doctrine. The authority to decide belongs to the person the parties chose, and it cannot be handed to a machine.


Lnu v. Blanche places the same kind of accountability on counsel through the signature. The arbitrator who adopts a fabricated authority in an award and the advocate who files one in a memorial are each personally answerable. Neither can shelter behind the tool. Neither can shelter behind a junior. The two decisions sit on opposite sides of the same tribunal and state the same principle.


That principle travels with the person, not the forum. The duties that bound Sethi and Rounds before the Ninth Circuit are the duties of competence, diligence, and candor that bind every advocate who appears before an arbitral tribunal. Many of the arbitrators reading this also sit as counsel. The professional obligation does not switch off when the proceeding moves from a courtroom to a hearing room.


What Lnu tells every advocate and arbitrator


  • Read what you cite. The duty of verification is absolute and indifferent to who drafted the text. If your name is on it, you have read and confirmed every authority in it.

  • Disclose fast and disclose fully. A hallucination caught early and disclosed plainly is a correctable error. A hallucination concealed, renamed, or denied becomes a candor violation, and the candor violation is what draws the suspension.

  • Do not delegate the checking. Not to a paralegal, not to an associate without a license, and least of all to the model that produced the problem.

  • Use the tool, own the output. Neither Lnu nor ARIHQ bans AI. Both hold that the human who signs or who decides carries the full weight of what the tool produces.


A Final Word

The hallucination story used to be the domain of the careless few in the trial courts. After June 3, 2026, that telling no longer holds. The standard is published, binding, and precise about where responsibility lies. Not when the model drafts. When the lawyer signs.


References

Comments

© 2026 TechCred. Design by Maas Creatives. Powered by Wix

bottom of page