Does an AI-hallucinated immigration brief trigger sanctions?

Short answer: yes. In Lnu v. Blanche, No. 24-4790, the Ninth Circuit disciplined two Orange County immigration lawyers after briefs in a petition for review of a BIA order contained nonexistent Ninth Circuit authorities and misattributed quotations. The sanctionable act was not “using AI” in the abstract. It was signing and filing briefs that vouched for authorities counsel had not personally verified. Last verified: Aug. 2, 2026, UTC. This Risk Digest record is for legal-risk reporting and filing hygiene; it is not legal advice. Because the primary Ninth Circuit PDF was not directly retrievable for this build, quoted language should be rechecked against the court’s order before reuse in a brief, sanctions response, or published quotation. The case facts and quoted passages below are confirmed through Metropolitan News-Enterprise and Reason’s Volokh Conspiracy reproductions of the order. [1][2]

What entered the record

The order’s center is narrow and concrete: two cited Ninth Circuit decisions did not exist. The fabricated authorities were identified as Eduardo v. Garland, 28 F.4th 742 (9th Cir. 2022) and Lay v. Holder, 729 F.3d 962 (9th Cir. 2013). The brief also attributed quotations to real immigration decisions, including Kamalthas v. INS, 251 F.3d 1279, 1284 and, as reported in the secondary reproductions, Avendano-Hernandez v. Garcia, 800 F.3d 1072, 1080; the spelling and citation of the latter should be checked against the primary order before republication. [1][2]

Those are not cosmetic defects. A fabricated appellate citation sends court staff, judges, and opposing counsel on a search for law that is not there. A misattributed quotation is worse than an ordinary miscitation because it gives the court a sentence with an institutional pedigree it never earned. In immigration litigation, where the petition may concern removal consequences and the record may already be procedurally compressed, that kind of false authority does not become less serious because it passed through a drafting chain.

The later Motion to Correct did not help. The Ninth Circuit’s order, as reproduced by the secondary sources, treated the lawyers’ characterization of the problem as “typographical” as part of the misconduct picture, not as an adequate correction. Calling nonexistent cases and invented language typographical errors shrinks the problem at the exact moment candor needs to expand. [1][2]

The sanction package was personal, public, and continuing

The Ninth Circuit suspended Mike Singh Sethi and William Rounds from practice before the court for six months and fined each lawyer $2,500. The order also required service of the disciplinary order on clients, opposing counsel, and presiding judges in every pending case, referred the matter to the California State Bar, and imposed a two-year firm-wide obligation in future filings to submit a sworn statement addressing whether generative AI was used, identifying the tool if one was used, and certifying personal review of citations and quotations. [1][2]

That last remedy is easy to underread. It is not a general Ninth Circuit rule requiring every lawyer in every case to disclose AI use. It is a case-specific, firm-wide remedial condition imposed after the court found signed filings containing nonexistent and misquoted authority. The order does not announce a universal ban on generative AI. It does show how quickly a false citation problem can become an office-wide disclosure burden once the court concludes that the lawyer’s review process cannot be trusted.

Where the violation attached

The most useful sentence for litigators is the court’s rule of attachment: the violation occurs at signing and filing. The signature is the lawyer’s personal attestation that the filing has been reviewed, including the legal authorities and quotations being offered to the court. Once the signed brief enters the docket, the lawyer cannot move responsibility backward into the drafting process and leave it there.

Reason’s reproduction quotes the court this way: “If an attorney files a brief with cases or quotations that do not exist, then he violates the rules. It generally does not matter if he pulled the hallucination, or even the entire brief, from the output of an artificial intelligence tool or from his own natural intelligence.” That passage should be checked against the official PDF before being quoted in litigation, but its rule is consistent across the available reproductions: AI provenance is not the defense because the filing signature is the act the court is regulating. [2]

Delegation received the same treatment. The order, as reproduced, states that it was no excuse that Sethi entrusted substantive cite-checking to subordinates or that he purportedly did not know subordinates had used generative AI. [1][2] That is not an anti-staff rule. It is the ordinary litigation rule applied to a new source of bad text: staff may draft, gather records, format tables, and run checks, but the lawyer whose signature goes on the brief owns the legal assertions the brief makes.

The distinction matters because many AI-filing discussions wander into tool debates too quickly. A court can be concerned about generative AI and still sanction the human act of filing a false brief. Conversely, a lawyer can avoid using AI entirely and still violate the same duty by filing a brief with invented cases, fabricated quotations, or unverified citations supplied by anyone else.

What discovery of a hallucination requires

The order also speaks to the moment after discovery. An attorney who learns that a filed brief contains hallucinated authority must notify the court and opposing counsel immediately, describe the error, and disclose how it came about. [1][2] A correction that obscures the nature of the defect risks becoming a second problem. If the case never existed, say that. If quoted language is not in the cited decision, say that. If AI output, a contract writer, an associate, or a shared template introduced it, the court’s concern is not satisfied by replacing the citation and moving on.

The practical consequence is uncomfortable but not complicated. Once the defect is known, the lawyer must stop treating the issue as a private workflow failure and treat it as a court-record problem. Opposing counsel has relied on the filing. Chambers may have reviewed it. The client may be exposed to delay, credibility damage, or additional motion practice. Those burdens are why the notice obligation belongs near the front of any remedial response, not after an internal postmortem.

Immigration-policy context should not be confused with the holding

Some readers will reach this record through immigration-policy searches, including queries around the U.S. visa bond program, permanent policy debates, and the $20,000 bond figure. That policy environment helps explain why immigration practitioners, legal reporters, and regulators are watching filing-risk developments closely in 2026. It does not change what Lnu v. Blanche held. The sanctioned filing was an immigration petition for review, but the misconduct was not a visa-bond-policy mistake. It was the submission of nonexistent and misquoted legal authority under attorney signatures.

The same restraint applies to broader AI-risk numbers. The Ninth Circuit’s order reportedly cited the Stanford legal-AI hallucination study by Magesh and coauthors, including figures that Westlaw and Lexis legal AI tools hallucinated in 17% and 33% of answers, respectively. Those figures are useful risk context, and they should be confirmed against the Stanford paper before publication as standalone statistics. They are not the basis of discipline in this case. The discipline rests on the filed brief, the false authorities, the inadequate correction, and the lawyer’s nondelegable signature obligation. [1][2]

What this record should change before the next filing

For a lawyer preparing a Ninth Circuit filing, the operational lesson is not to write a firm AI manifesto. It is to make sure no case, quotation, pincite, or parenthetical reaches the signature block without personal verification by the lawyer who will sign. If drafting was delegated, the final review still has to identify who checked the cited authority, where it was checked, and whether quoted language appears in the source being represented to the court.

For EOIR-specific questions, this record should sit beside the companion Regulation & Ethics record on EOIR PM 25-40, 8 C.F.R. § 1003.102 discipline grounds, the absence of a blanket national AI ban, and the role of local standing orders. For pre-filing operations, it should later connect to the Verification Workflows checklist. But Lnu v. Blanche itself is already enough to answer the filing-risk question: in the Ninth Circuit, an AI-hallucinated immigration brief can trigger attorney discipline because the court treats the signed filing as counsel’s personal representation that the law cited in it exists and says what the brief claims it says.

References

1. Two O.C. Attorneys Suspended Relating to AI Fabricated Cites, Metropolitan News-Enterprise, June 4, 2026.
2. Ninth Circuit on AI Hallucinations, Reason / Volokh Conspiracy, June 3, 2026.