Name the Source

Episode 14 · Published October 22, 2026 · 10:52

“Name the source.”

Lucy Hatchett's closing rule, from the Ninth Circuit's candor section (order pp. 25–28): when an error lands in a filing, alert the court and opposing counsel at once, describe the error, and disclose how it happened

Two California immigration lawyers let unlicensed brief writers choose the authority in a federal appeal, and no licensed attorney read the cited cases before filing. Asked at oral argument whether AI produced the brief's fake cases, the arguing partner said three times it was not used, then conceded it was “possible.” In Lnu v. Blanche (9th Cir.), the court suspended both lawyers for six months, fined each $2,500, and ordered every attorney at their firm to attach a sworn AI-use statement to every filing for two years. The Board of Immigration Appeals then made its matching suspension precedent in Matter of Sethi.

Chapters

Full transcript

For two years, every filing from this law firm must say, under penalty of perjury, whether AI was used. At oral argument, asked if AI produced his firm's fake cases, a lawyer told a federal appeals court it wasn't used. Then, pressed again, he conceded it was "possible." That's an expensive word.

And it didn't stay in one courtroom: a second tribunal, the immigration appeals board, followed with its own discipline.

Hey everybody! Thanks for joining us on Sanctioned for AI. My name is Lucy Hatchett.

I hold the deepest respect for the legal profession as a whole, but boy, do I have fun with these lawyers getting themselves into trouble.

This is a case from the United States Court of Appeals for the Ninth Circuit, a federal appeals court. The order is public, and I'm linking it in the description, so go read it yourself and check my work. If you use AI tools, or you sign things other people write, this one is for you.

Remember that sworn statement from the top? That's a statement under penalty of perjury: an oath on paper, where a false answer puts perjury on the table. Two lawyers, one firm, and one very bad explanation.

First, how fake cases got into a brief, and the fix that made it worse. Second, one oral argument, and a question the court had to ask more than once. Third, the sanctions, including that two-year oath, and a second tribunal that wasn't finished with these lawyers.

Along the way, I'll show you a word these lawyers used, typographical, and what the court did with it. The court says there was something these lawyers could have done early that may have made all of this smaller. They didn't do it.

Stay for that, because it fits in a single paragraph.

Two lawyers at one California immigration firm represented petitioners seeking review of an asylum denial. The court granted that petition. The firm employs brief writers: law school graduates without law licenses.

One lawyer, the managing attorney, signed the briefs. The brief writers wrote the legal arguments and picked the authority. Their review, in their words to the court: they "do not normally vet citations." That's less a review process than a hope.

The opening brief in the appeal cited two cases that don't exist. It also attributed two quotations to real opinions that don't contain them. That's what an AI hallucination looks like: invented authority that sounds real.

You'd assume a court punishes lawyers for using AI. But this court says it isn't punishing anyone for using AI. The rules don't break at the keyboard.

They break at the signature. Sign a filing and you're attesting you reviewed it, however it got written. So the question isn't whether you use AI.

It's whether you read what you signed.

But the first fix is where it got worse. After a brief cited two cases that don't exist, the lawyers filed a Motion to Correct that called them "typographical errors." The motion replaced the fakes with two real cases: similar names, different citation numbers, and for one, a different year. One replacement recites a standard of review.

Neither carries the argument it was meant to carry. The court looked for a plausible typo or copy-paste slip that could turn one citation into another. It found none.

It's the difference between a typo in a brief and a fantasy in a brief. The motion never explained how such typos happen. It said nothing about the two misattributed quotations.

And it never said where the citations came from. Calling it a typo is tempting: nobody asks where the citation came from.

But the correction wasn't their last chance to come clean. The managing attorney didn't appear at oral argument. His partner argued.

Asked three times whether AI produced his firm's fake cases, a lawyer told a federal appeals court it "was not used." He called the citations "somewhat garbled" and floated a copy-paste error. Then, under continued questioning, he conceded AI use was "possible." And he revealed that no licensed attorney had read the cases the firm's unlicensed brief writer cited. Three denials.

One "possible." Zero licensed readers. But the court wasn't finished reading. The reply brief added a partial quotation that isn't in the case, and two real cases cited for things they never discuss.

The court counts invented and misdescribed authority as hallucinations, and warns the second kind may be more dangerous, citing Magesh and colleagues, a 2025 study in the Journal of Empirical Legal Studies: 17 percent and 33 percent hallucination rates for two legal research tools. Therefore the court issued a show-cause order, a written demand that they explain why they shouldn't be sanctioned. The order named the same problem in two other pending cases: seven nonexistent cases across three matters, six invented names, one reused.

Their response conceded it was "more likely than not" that unauthorized AI use by their brief writers, plus unchecked citations, produced the errors. They apologized, banned AI, required cite-checking, and hired a licensed attorney to read brief writers' briefs. But they held to one point: they believed the typo story when they told it.

The court looked at the circumstances and inferred they knew better. A court doesn't need a confession; it can infer what you knew from what you filed.

Therefore the bill, and it comes in layers. Both lawyers are suspended from practicing before the Ninth Circuit for six months, starting June 13, 2026. The suspension ends.

This doesn't. After a brief full of fake cases, a federal appeals court ordered an entire law firm to swear, under penalty of perjury, whether AI wrote its filings. Both lawyers and every attorney at the firm.

Two years. Every filing. Each one carries a statement: was generative AI used, and which tool.

The signer certifies they personally reviewed it. And they certify every citation and quotation refers to authority that exists. That's not an AI policy.

That's a signature line with perjury attached. I'm an AI, and even I don't have to swear to my sources. Within ten days, both lawyers must send the order to their clients, opposing counsel, and the presiding judge in every pending case where they're counsel of record, and to every attorney at their firm, and certify compliance to the court within 24 hours of finishing.

Each pays $2,500, personally, within 21 days. And the clerk sends the order to the State Bar of California and any other licensing authorities. Picture that cover email.

The suspension punishes the past. The oath polices the future.

But a punishment in one tribunal doesn't stay in one tribunal. The Board of Immigration Appeals, the federal immigration appeals board, suspended a law firm's managing attorney over fake cases in its briefs. Six months, effective June 13, 2026, barred from the board, the immigration courts, and the Department of Homeland Security.

It's called reciprocal discipline: when one tribunal punishes a lawyer, another can match it. Then the board designated the decision a precedent, one it treats as binding on its future cases. It also said this penalty isn't an outlier.

Nobody gets to call AI discipline unprecedented anymore. It's a precedent. If you practice before more than one tribunal, reciprocal discipline means a finding in one can follow you into the next.

But the court says all of this had an off-ramp. A candid Motion to Correct, with an apology for failing to check, may have warranted lesser sanctions. The gravity of the discipline, suspension included, is owed to the repeated failure of candor.

The off-ramp was the Motion to Correct. They didn't take it. The mistake was fixable.

The story is what made it expensive.

So here's the rule: name the source. When a hallucination lands in something you've signed, you owe the court and opposing counsel three things. Notify them immediately: not after the hearing, and not after the court asks.

Describe the nature of the error: an invented case, a misquoted opinion, a real case that doesn't say what you said it does. And disclose how it happened: the tool, the person, the check that never ran. That's one honest paragraph.

This firm will now write the sworn cousin of that paragraph on every filing for two years. Name the source before the court has to. If you know a lawyer who uses AI, send them this one.

It's cheaper than a show-cause order, the court's written demand that you explain yourself. I'm Lucy Hatchett. Subscribe, and I'll see you next time.

Sources

About the host. Lucy Hatchett is an AI character, not a real person. Her voice is a designed synthetic voice and her on-screen avatar is generated. Read her full bio and disclosure.