top of page
Office Collaboration Scene

The Ninth Circuit Polite Eulogy to AI

AI Rules Have Arrived

​For two years the question hanging over every brief written with help from a chatbot was: when does this get a lawyer sanctioned, exactly? Trial courts had been answering it case by case, sometimes for $250, sometimes for $110,000, with no unifying logic beyond "you should have read your own brief." On June 3, the Ninth Circuit published the first appellate framework, and it is nothing the careless will enjoy.

​

The case is Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026). The facts are almost too on-the-nose. Two California immigration attorneys at the same firm let an unlicensed law-school graduate — the firm's job title for him was "Brief Writer" — draft an opening brief with generative AI. Nobody checked the work. The brief went to the panel containing multiple nonexistent cases, misattributed quotations, and "gross misrepresentations of real cases." When the court noticed, the attorneys filed a Motion to Correct calling the fabricated citations "typographical or copy-paste errors." A typo, apparently, is now what we call a hallucinated federal case.

The panel — Judges Paez, Bea, and Forrest — was not amused.

The Sanctions

Each attorney walked away with:

  • $2,500 monetary sanction under Federal Rule of Appellate Procedure 46(b) and Ninth Circuit Rule 46-2.

  • Six-month suspension from practice before the Ninth Circuit, starting ten days after the order.

  • Referral to the State Bar of California and any other licensing authority that wants a turn.

  • A requirement to personally hand the order to every client, every opposing counsel, and every presiding judge in every pending state or federal matter — within ten days.

  • And the kicker: a two-year sworn AI disclosure regime binding the entire firm. Every future filing by any attorney at the firm must include a statement, under penalty of perjury, identifying whether generative AI was used, naming the tool, and certifying that the signing attorney has personally read the filing and confirmed that every citation refers to existing authority. This requirement is firm-specific. For now.

That last piece is the real precedent. The court built a template that any federal judge in the Ninth Circuit can pull off the shelf the next time a similar brief crosses their desk.

The Four-Rule Holding

The opinion distills into four rules, and these are the lines that will appear in CLE slides for the next decade.

1. Using AI is not the violation. Signing and filing verification is. "[T]he rules are not violated at the point of research and drafting, but at the point of signing and filing." The panel was emphatic: "We do not sanction Sethi and Rounds for the simple fact that they or their subordinates used generative AI." You can draft with whatever tool you want. You just cannot put your name on something you have not read.

2. Verification means actually reading the case. "Before filing any court document, an attorney must carefully check every case citation, fact, and argument to make sure that they are correct and proper. Attorneys cannot delegate that role to AI, computers, robots, or any other form of technology." The court's standard is not "did you run a citator." It is "did you read the authority and reason about whether it supports the proposition." Those are not the same task. A KeyCite green light tells you a case still exists. It does not tell you the case stands for what your brief claims.

3. When you find an error — including a hallucination — disclose it immediately and disclose the source. "When an attorney learns of any error in a filing — including generative AI hallucinations — he should immediately alert the court and opposing counsel of the error and disclose its source." Calling a fabricated case a typo is itself a separate violation of the duty of candor. The cover-up, as ever, is worse than the crime.

4. Sanctions track candor, not AI use. This is the part of the opinion that should be tattooed on the inside of every litigator's wrist. The panel said: "If, in the Motion to Correct, the attorneys had disclosed that AI was used in the opening brief against firm policy and apologized for failing to check the brief, lesser sanctions may have been warranted." The six-month suspension was not the price of using ChatGPT. It was the price of telling a federal court of appeals that a fake case was a copy-paste error.

​

The Ninth Circuit, in other words, has done what good appellate courts do: it has told the bar exactly how to avoid the worst version of this outcome the next time it happens. And it will happen.

A Brief Aside on Federal Cameras

While the Ninth Circuit was writing the AI rulebook, the federal judiciary was using AI to argue against something else entirely. On June 18, the Senate Judiciary Committee advanced two bipartisan bills — the Sunshine in the Courtroom Act and the Cameras in the Courtroom Act — that would let federal trial and appellate judges allow television cameras in their courtrooms and would require the Supreme Court to broadcast oral arguments. The judiciary has opposed cameras for forty years on classical grounds (witness security, juror intimidation, the dignity of the proceedings). What is new is the Judicial Conference's letter to the committee chair warning that deepfake recordings of court proceedings could cause "serious and irreparable reputational harm" and be weaponized for fraud.

That argument matters less for the legislation, which will limp along on its usual schedule, and more for what it implies about evidentiary practice. If the federal judiciary is officially worried about deepfaked recordings of its own proceedings, the day is close when one of us will receive video discovery that may or may not have been generated. Proposed Federal Rule of Evidence 707, on machine-generated evidence, and the proposed Rule 901(c) burden-shifting mechanism for challenged authenticity, are moving on a timeline that will land them in our practice well before 2028. Read them now. Read them again before your next deposition designation.

The Numbers Behind the Trend

Damien Charlotin's database tracks AI hallucination cases worldwide; he is currently logging well over 1,600 legal decisions, with several hundred resulting in sanctions and a growing list triggering bar referrals. The dollar figures have detached from the ceremonial. A short tour of the last twelve months:

  • Fifth Circuit, February 2026: $2,500 against attorney Heather Hersh in Fletcher v. Experian Information Solutions, No. 25-20086. Sixteen fabricated quotations and five additional serious misrepresentations of law or fact in a single reply brief. The court noted she also appeared to have used AI in her response to the show-cause order. (Reuters)

  • Southern District of Ohio, February 2026: $7,500 plus a contempt finding against two attorneys in Kettering Adventist Healthcare v. Collier, No. 25-273, for filings containing more than a dozen fabricated or misrepresented citations — including a real case repeatedly cited for a proposition the opinion does not contain, even after the court flagged the misquote.

  • Sixth Circuit, March 2026: Whiting v. Athens and two companion appeals. The panel dismissed the appeals as frivolous, imposed $15,000 in punitive sanctions against each of two attorneys, and ordered them to pay opposing counsel's fees and double costs. With fees and costs, the total exposure runs above $116,000 jointly and severally. (Reuters; Dinsmore analysis.)

  • Q1 2026 aggregate, per EDRM tracking: more than $145,000 in monetary sanctions for AI-fabricated citations, including a $110,000 single-attorney penalty out of Oregon.

Courts are no longer startled. They are exasperated. The 5th Circuit panel that fined Hersh said the problem "shows no sign of abating." When an appellate court uses the phrase "shows no sign of abating," it is preparing the bench, and posterity, for the next ten years of escalating fines.

What This Means for Attorneys

If you read only one paragraph of this article, read this one.

​

The line is not "do not use AI." Every court that has looked at the question has said AI assistance in legal drafting is permitted. The line is: do not sign anything you have not personally read against the actual source. The verification standard is not citator-green. The verification standard is "I pulled the case, I read the cited passage, and I am satisfied that the authority supports the proposition for which I am citing it." If a junior associate, a contract attorney, a Brief Writer, or a chatbot drafted the paragraph, that is fine. The signature line on the brief is a representation to the court that you, personally, did the verification. There is no theory of vicarious AI liability. There is only signature liability.

​

Build the verification step into your workflow before the firm-wide AI disclosure order from your Ninth Circuit shows up in your in-box. Document it. If you use a legal AI tool with citator integration (Westlaw, Lexis, CoCounsel, vLex), do not assume the citator output is verification. The 5th Circuit's $2,500 sanction in Fletcher came down on a brief drafted with a commercial legal AI product. The commercial pedigree of the platform was no shield against the hallucination, and it did not mitigate the sanction.

​

And when something does go wrong — because at some point in the next two years something will go wrong, somewhere, in someone's brief — the Ninth Circuit has now told us the playbook. Disclose the error. Disclose the source. Apologize. Do not, under any circumstances, tell the court that a hallucinated case is a typo.

​

The AI rules have arrived.

bottom of page