MALKEET LNU v. Mike Singh Sethi; William Rounds, Interested Parties. (2026)

ORDER

Attorneys Mike Singh Sethi and William Rounds filed briefs in this Court with multiple nonexistent cases, misattributed quotations, and gross misrepresentations of real cases. Sethi and Rounds claimed that the errors were the product of innocent typographical mistakes. And they repeatedly denied the possibility that generative artificial intelligence (“AI”) might have produced the errors. Having identified other cases in which Sethi or Rounds filed briefs that presented similar problems, we ordered them to show cause why they should not be sanctioned, suspended, or disbarred from practice before this Court. We have considered their response, and we now impose discipline.

We have previously stricken a brief in its entirety where the brief was composed almost entirely of fabricated or inaccurate citations. Grant v. City of Long Beach, 96 F.4th 1255, 1257 (9th Cir. 2024). But we have yet to explain the proper course when the brief as a whole does not merit such a consequence, or what disciplinary consequences might be warranted in such a case. We issue this disciplinary order, and explain our reasoning at some length, as a warning to the members of this Court's bar: be aware of the risks of overreliance on generative AI, read everything cited in a court filing—whether drafted by generative AI or not—and disclose quickly and transparently generative AI hallucinations that are inadvertently included in court filings.

I. Background

A. Briefing and Motion to Correct

Sethi and Rounds are attorneys licensed to practice law in the State of California and before this Court. They are part of Sethi Law Group/US Legal Group, APC (“Sethi Law Group” or “the Firm”), a law firm that primarily handles immigration matters and is based in Orange County, California.  1 We assume, based on the Firm's website and Rounds's representations at oral argument, that Sethi is the managing attorney. Oral Argument at 12:53–12:59, Lnu, et al. v. Bondi (No. 24-4790), https://www.ca9.uscourts.gov/media/video/?20251020/24-4790/ [https://perma.cc/ZC5H-H26T].

Sethi and Rounds served as counsel for the petitioners in this case, Lnu, et al. v. Blanche, No. 24-4790. Petitioners sought review of an order of the Board of Immigration Appeals (“BIA”), which dismissed their appeal from an Immigration Judge's denial of their applications for asylum, withholding of removal, and protection under the Convention Against Torture (“CAT”). The petition for review, which we granted, principally concerned an adverse credibility determination and the sufficiency of independent documentary evidence. Lnu v. Bondi, No. 24-4790, 2025 WL 3079970 (9th Cir. Nov. 4, 2025).

Sethi filed an opening brief in this case with multiple fabricated citations and quotations. 2 Sethi cited two cases that do not exist and never existed: “Eduardo v. Garland, 28 F.4th 742 (9th Cir. 2022),” and “Lay v. Holder, 729 F.3d 962 (9th Cir. 2013).” 3 And Sethi twice attributed quotations to real opinions in which the quoted language does not appear: Kamalthas v. INS, 251 F.3d 1279, 1284 (9th Cir. 2001), and Avendano-Hernandez v. Lynch, 800 F.3d 1072, 1080 (9th Cir. 2015).  4 The Attorney General did not flag the fabricated citations in the answering brief.

After we denied the parties’ joint motion to submit Lnu on the briefs, Sethi filed a Motion to Correct the Record Re: Errata to Petitioner's Opening Brief (“Motion to Correct”). The Motion to Correct represented that the two nonexistent cases—“Eduardo v. Garland” and “Lay v. Holder”—were “typographical errors.” Sethi sought to replace those cases with two cases that have similar names, different reporter numbers, and in the case of “Lay,” a different year: Udo v. Garland, 32 F.4th 1198 (9th Cir. 2022), and Lai v. Holder, 764 F.3d 1098 (9th Cir. 2014). He also sought to “correct” the holding for which “Lay” was cited. Sethi did not explain how such significant typographical errors might have occurred. Nor did he address the quotations misattributed to Kamalthas and Avendano-Hernandez.

The replacement citations do not support their intended propositions. Sethi twice cited “Eduardo” for the proposition that “country conditions reports” can corroborate an applicant's fear of persecution. Udo, however, does not discuss country conditions reports or the issue of particularity in such reports. See 32 F.4th at 1202, 1205. The case mentions country conditions reports once, but only to distinguish a prior case, and only to say that the country reports in that case had “little probative value.” Id. at 1205. 5

As modified by the Motion to Correct, Sethi apparently intended to cite Lai, 764 F.3d at 1108, not “Lay,” for the proposition that “[t]he Ninth Circuit has held that corroboration from witnesses who cannot appear in person should not be discounted solely due to their absence.” 6 But Lai does not discuss the weight of testimony from absent witnesses. 7

B. Oral Argument

Sethi did not appear for oral argument. Rounds appeared on behalf of Petitioners instead. At oral argument, we asked Rounds to explain the errors identified in the Motion to Correct. Oral Argument at 8:47–9:00. Rounds stated that the intended citations were “somewhat garbled” and reiterated the claim that Sethi had intended to cite real cases. Id. at 9:05–9:14. Rounds claimed that the real cases “stand for the same proposition.” Id. at 9:14–9:20. As for the source of the errors, Rounds asserted that he was “not sure” but that “it looks like it was a copy and paste error or something like that.” Id. at 9:05–9:28.

We then asked Rounds whether the errors might have been the product of generative AI, to which Rounds said that AI “was not used.” Id. at 9:28–9:31. Rounds explained that the quotation misattributed to Kamalthas appeared in a different case, but did not explain how the misattribution occurred. Id. at 9:33–10:00. We again asked Rounds whether generative AI might have been used to supplement the briefs, to which Rounds again said: “No. AI was not used ․” Id. at 10:05–10:16. We raised the other quotation misattributed to Avendano-Hernandez and again asked whether the error was the product of generative AI, to which Rounds again said “No.” Id. at 10:17–45.

After further questioning on the source of the errors, Rounds finally conceded that it was “possible” that AI might have been used by the individual who drafted the briefs. Id. at 10:59–11:11. He clarified that although Sethi's name was on the briefs, Sethi did not draft the briefs, but only “reviewed” them. Id. at 12:28–12:40. Rounds later revealed that the brief writer was not yet licensed to practice law, id. at 13:01, and that no licensed attorney read the cases cited by the unlicensed brief writer, id. at 25:11–25:38.

C. Order to Show Cause & Response

After oral argument, we ordered Sethi and Rounds to show cause (“Order to Show Cause”) why they should not be sanctioned, suspended, or disbarred from practice before this court for “conduct unbecoming a member of [this] court's bar,” Fed. R. App. P. 46(b)(1)(B), and for “violating applicable rules of professional conduct,” 9th Cir. R. 46-2(a). No. 24-4790, Dkt. 41.

In the Order to Show Cause, we identified additional issues in the Lnu reply brief. Part of the quote attributed to Bandari v. INS, 227 F.3d 1160, 1167 (9th Cir. 2000), does not appear in that opinion. 8 Sethi cited Singh v. Gonzales, 491 F.3d 1090, 1096–97 (9th Cir. 2007), for a proposition related to adverse credibility, but the case does not discuss adverse credibility. Sethi represented the holding of Zahedi v. INS, 222 F.3d 1157, 1165 (9th Cir. 2000), as concerning affidavits, but the opinion does not discuss affidavits.

We also identified similar issues in briefs filed by Sethi in other cases pending in this Court. In the opening brief for Anguiano Alvarado v. Bondi, No. 25-2485, Sethi cited three cases that do not exist: “Garcia v. Garland, 60 F.4th 1239 (9th Cir. 2023)”; “Hernandez v. Sessions, 873 F.3d 1120 (9th Cir. 2017)”; and “Vasquez-Zavala v. Garland, 69 F.4th 1086 (9th Cir. 2023).” In the opening brief for Contreras Pelayo v. Bondi, No. 24-5168, Sethi cited two cases that do not exist: “Eduardo v. Garland, 28 F.4th 742 (9th Cir. 2022)” and “Gutierrez-Alm v. Garland, 72 F.4th 446 (9th Cir. 2023).” 9

Sethi and Rounds timely filed a joint response to our Order to Show Cause with individual declarations (collectively, “Response”).  10 In the Response, Sethi and Rounds first explained the Firm's brief writing process. Sethi Law Group employs law school graduates, who are not yet licensed attorneys, to write briefs (“Brief Writers”). While the statement of facts for each brief is prepared by an attorney, the Brief Writers are left to write the legal arguments and identify supporting authority for those arguments. Sethi and Rounds explain that they “do not normally vet citations used by the Brief writer during [their] review.”

Sethi and Rounds also explain why, at the time Sethi filed the Motion to Correct and Rounds appeared at oral argument, they believed that the errors were not the product of generative AI but innocent typographical mistakes. Rounds discovered the “Eduardo” and “Lay” errors while preparing for oral argument. Rounds then notified Sethi. They assumed the irregularities were innocent typographical errors, either from copying and pasting only part of a citation from a prior brief, or from incorrectly typing in a citation from the firm's legal research service or a prior brief. Neither of them suspected the use of generative AI because the Firm prohibited its use and required the Brief Writers to check all their citations.

To support the assertion that they innocently believed the “Eduardo” and “Lay” errors were typographical, Sethi and Rounds state that “it was believed that the replacement citations provided support [for the claims in the briefs], however marginal, and were the ones originally intended.” Rounds argues that Udo supports the proposition that “country conditions” “can corroborate an applicant's claims ․ even if they do not mention the applicant by name.” To that end, Rounds characterizes Udo as reaching an analogous holding: the agency there should have considered written statements from Udo's family members even though the family misidentified the name of the hotel where Udo was persecuted. This is a gross mischaracterization of Udo. 11

As for Lai, Rounds argues that it indirectly supports the claims in the brief because Lai reiterates the general “totality of the circumstances” rule in adverse credibility determinations. But Sethi cited “Lay” for a specific holding, not a general standard.

Notwithstanding these explanations and arguments, Sethi and Rounds now concede it is “more likely than not” or “probable” that the errors were the product of unauthorized AI use by their Brief Writers and a failure to cite check. They apologize to the Court and its staff and note their embarrassment. And they note that they have hired a licensed attorney to read all briefs prepared by the Brief Writers in future cases.

Sethi and Rounds also stated their intention to remedy similar errors in other pending matters. In at least one of those cases, Sethi filed a motion to correct that did not identify the original citations as fabricated or the product of generative AI, but simply as “errors.” See, e.g., No. 24-5168, Dkt. 21.

II. Authority to Impose Discipline

We have the authority to impose discipline authorized by rule and statute. See Chambers v. NASCO, Inc., 501 U.S. 32, 40, 111 S.Ct. 2123, 115 L.Ed.2d 27 (1991). Federal Rule of Appellate Procedure (“FRAP”) 46(b)(1)(B) authorizes us to suspend or disbar attorneys from practice in this court for “conduct unbecoming a member of the court's bar.” Ninth Circuit Rule 46-2(a) authorizes us to “impose discipline on any attorney practicing before this Court who engages in conduct violating applicable rules of professional conduct, or who fails to comply with rules or orders of this Court.” Because Sethi and Rounds maintain their principal office in California, we apply the California Rules of Professional Conduct and California's State Bar Act as the “applicable rules of professional conduct.” 9th Cir. Gen. Ord. 12.9(a); 9th Cir. R. 46-2(a); Cal. R. Pro. Conduct 1(b)(2).

Our Circuit Rules authorize discipline consisting of “disbarment, suspension, reprimand, counseling, education, a monetary penalty, restitution, or any other action that the Court deems appropriate and just.” 9th Cir. R. 46-2(a).

III. Discussion

Sethi and Rounds violated multiple rules of appellate procedure and professional conduct. Sethi did so when he signed and filed briefs in this Court with nonexistent cases, misattributed quotations, and gross misrepresentations of real cases. The contentions associated with these falsities lacked “citations to the authorities,” Fed. R. App. P. 28(a)(8)(A), and Sethi's conduct violated his duties of competence and diligence, Cal. R. Pro. Conduct 1.1, 1.3, as well as his duty to present meritorious claims, id. 3.1(a)(2). Sethi and Rounds also violated their duty of candor when they represented the errors as innocent typographical mistakes and when they affirmatively denied that generative AI might have been the source of the errors. Cal. R. Pro. Conduct 3.3(a)(1). We detail each violation in turn.

A. Generative AI

We first clarify what this order is not about. We do not sanction Sethi and Rounds for the simple fact that they or their subordinates used generative AI. The Federal Rules of Appellate Procedure and our Circuit Rules “do not directly regulate the process by which a party or an attorney produces [a] filing—such as writing it personally with no assistance, delegating part of its preparation to a subordinate, or employing generative artificial intelligence.” 9th Cir. R. 32-1, Advisory Comm. Note (Dec. 1, 2025) [“Rule 32-1 Adv. Comm. Note”]. California's State Bar Standing Committee on Professional Responsibility and Conduct has acknowledged that “[g]enerative AI is a tool that has wide-ranging application for the practice of law,” and has published a guide on how to use responsibly generative AI consistent with the Rules of Professional Conduct (“Cal. Bar Practical Guidance”). State Bar of Cal. Standing Comm. On Prof. Resp. & Conduct, Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law 1 (2025), https://www.calbar.ca.gov/sites/default/files/2025-09/Generative-AI-Practical-Guidance.pdf [https://perma.cc/UAS5-8XXP]; see also Noland v. Land of the Free, L.P., 114 Cal. App. 5th 426, 446, 336 Cal.Rptr.3d 897 (2025) (“[T]here is nothing inherently wrong with an attorney appropriately using AI in a law practice.”).

While not inherently unethical or irresponsible, using generative AI without rigorously checking its output does present a higher risk of violating certain ethical and procedural rules. This is due to the nature of mistakes that generative AI is prone to make, at least in its current stage of advancement. Two types of mistakes, or “hallucinations,” are most relevant: fabrications and inaccuracies. Magesh et al., Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools, 22 J. Empirical Legal Stud. 216, 221 (2025). Fabrications are instances in which the generative AI tool provides cases or quotations that do not exist at all. Id. at 221, 230. Inaccuracies are more subtle. Id. at 222. The generative AI tool might cite to real authorities but provide an answer that is legally or factually inaccurate or not supported by the citation. Id. at 221–22. 12

Fabrications are the most notorious hallucinations, but inaccuracies may prove more dangerous to our profession in the long run. Inaccuracies are more likely to go unnoticed by attorneys and judges because they are not always susceptible to facial checks. Rather, “[i]dentifying these misunderstandings often requires close analysis of cited sources.” Id. at 225. With close analysis, some inaccuracies might be clear—for instance, claiming that a case expressly stands for a proposition about a certain topic when the case does not discuss that topic at all. Others may be difficult to distinguish from poor legal reasoning.

And inaccuracies are common, even in newer generation models that produce fewer fabrications. Including inaccuracies, legal-specific generative AI tools from Westlaw and Lexis hallucinated 17% and 33% of answers, respectively, to a representative set of queries run in 2024. 13 Id. at 222–23, 225. The most common error modes of the latest generation tools include misunderstanding holdings, failing to distinguish between legal actors (e.g., presenting a rejected party argument as the holding of the court), and failing to respect the hierarchy of authorities. Id. at 225–30. In other words, the sort of errors that we might expect a first-semester law student to make, but certainly not licensed attorneys appearing before this court.

As we will explain, filing briefs with hallucinated fabrications and inaccuracies violates procedural and ethical rules. Lawyers using generative AI must thus be aware of the tendency of generative AI to make these mistakes and guard against them. See, e.g., Noland, 114 Cal. App. 5th at 446, 336 Cal.Rptr.3d 897 (“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.” (citation modified)).

It is ultimately irrelevant to the disciplinary analysis (except for the duty of candor), however, whether Sethi, Rounds, or anyone at the Firm actually used generative AI or was aware of its tendency to make mistakes. Attorneys do not need cutting-edge technology to fabricate citations and make demonstrably false and unsupported statements. And the ethical and procedural rules we apply today do not turn on the source of such error. That is, the rules are not violated at the point of research and drafting, but at the point of signing and filing. If an attorney files a brief with cases or quotations that do not exist, or completely misrepresents what a real authority stands for, it generally does not matter if he pulled the hallucination or misrepresentation from the output of an artificial intelligence tool or from his own natural intelligence.

However legal papers are prepared, and however legal technology develops, our procedural and ethical rules apply with equal force. Just as faithful adherence to those rules would prevent the submission of generative AI hallucinations, such adherence would also prevent the submission of similar human-generated errors. Sethi and Rounds failed to adhere to the rules. It is for those failures that we impose discipline, as discussed below.

B. Filing

Sethi violated the Federal Rules of Appellate Procedure and the California Rules of Professional Conduct when he filed briefs with nonexistent cases, misattributed quotations, and gross misrepresentations of real cases.

1. Applicable Rules

The Federal Rules of Appellate Procedure provide that “contentions” must be supported “with citations to the authorities ․ on which the [party] relies.” Fed. R. App. P. 28(a)(8)(A); id. 28(b). FRAP Rule 32(d) in turn provides that “[e]very brief, motion, or other paper filed with the court must be signed by the party filing the paper or, if the party is represented, by one of the party's attorneys.”

Our Circuit Advisory Committee recently explained that the signature requirement of FRAP 32(d) is our Court's primary guard against inaccurate filings. “Regardless of how the filing is prepared, the signature is an attestation that the signer has reviewed the filing and is responsible for the accuracy of its contents.” Rule 32-1 Adv. Comm. Note. In other words, no matter how a filing is prepared—whether by delegation, partnership, solo work, or generative AI—the party or attorney who submits the filing with this Court must read the authorities cited therein to ensure that they are real, properly attributed, and accurately represented.

The California Rules of Professional Conduct impose duties of competence and diligence. Cal. R. Pro. Conduct 1.1, 1.3. These duties apply to attorneys who prepare filings with generative AI and review such filings. Cal. Bar Practical Guidance 2–3 (Duties of Competence and Diligence). At a bare minimum, these duties require lawyers to review the output of generative AI for fabrications and basic accuracy: Are these cases and quotations real? Does this case discuss the issue at hand?

But a competent and diligent attorney must do more than prompt generative AI, check that the citations provided by the AI are real and the subject matter roughly on point, and call it a day. Cal. Bar Practical Guidance 3 (Duties of Competence and Diligence) (“The duty of competence requires more than the mere detection and elimination of false AI-generated results.”). A competent and diligent attorney must also read and reason. The California Court of Appeal put it well: “To state the obvious, it is a fundamental duty of attorneys to read the legal authorities they cite in appellate briefs or any other court filings to determine that the authorities stand for the propositions for which they are cited.” Noland, 114 Cal. App. 5th at 445, 336 Cal.Rptr.3d 897 (original emphasis). This exercise of professional judgment “cannot be delegated to generative AI and remains the lawyer's responsibility at all times.” Cal. Bar Practical Guidance 3 (Duties of Competence and Diligence). That is, there is no substitute for “critical attorney analysis fostered by traditional research and writing.” Id.

The duties of competence and diligence also apply, of course, to attorneys who do not use generative AI and do not suspect its use in the drafts they review. And the substance of the duties is the same. “Simply stated, no brief, pleading, motion, or any other paper filed in any court should contain any citations—whether provided by generative AI or any other source—that the attorney responsible for submitting the pleading has not personally read and verified.” Noland, 114 Cal. App. 5th at 430, 336 Cal.Rptr.3d 897 (original emphasis).

The California Rules of Professional Conduct also provide that “[a] lawyer shall not ․ present a claim or defense in litigation that is not warranted under existing law, unless it can be supported by a good faith argument for an extension, modification, or reversal of the existing law.” Cal. R. Pro. Conduct 3.1(a)(2). A legal argument supported only by hallucinations is neither “warranted under existing law” nor a “good faith argument,” Cal. R. Pro. Conduct 3.1(a)(2). That is, “[a] fake opinion is not ‘existing law’ and citation to a fake opinion does not provide a non-frivolous ground for extending, modifying, or reversing existing law, or for establishing new law.” Noland, 114 Cal. App. 5th at 445, 336 Cal.Rptr.3d 897 (quoting Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 461 (S.D.N.Y. 2023)).

2. Sethi's Conduct

Sethi violated these procedural and ethical rules when he cited cases that do not exist. The claims that those cases were meant to support were presented without “citations to the authorities,” Fed. R. App. P. 28, because a fabricated citation is not an authority. And the claims were neither supported by existing law nor good faith arguments. Cal. R. Pro. Conduct 3.1(a)(2); Noland, 114 Cal. App. 5th at 445, 336 Cal.Rptr.3d 897.

It is irrelevant that other cases may stand for the propositions asserted. If other cases supported Sethi's “contentions,” it was his responsibility to cite them. Fed. R. App. P. 28(a)(8)(A); see also Noland, 114 Cal. App. 5th at 447, 336 Cal.Rptr.3d 897 (“It is counsel's job—not this court's—to identify legal authority to support appellant's contentions. The existence of (uncited) cases in support of plaintiff's legal contentions does not excuse the fraudulent case cites.” (original emphasis)).

Nor do we see how it might matter, as urged in the Response, that the citations were purportedly offered for points that were “not regarded as case determinative, or hotly contested.” Insertion of useless bulk in a brief serves only to distract the Court and perhaps pad the client's bill. And all claims in a brief, no matter how minor, must be made with citations to real authority or a good faith argument for a change in the law.

Sethi also violated these rules when he attributed quotations to opinions in which the quoted language does not appear. For instance, Sethi cited Kamalthas, 251 F.3d at 1284, for the following proposition: “An adverse credibility determination is not necessarily a death knell to CAT protection.” This quotation does not appear in Kamalthas. The substance of the quotation is thus a “contention[ ]” provided without “citation[ ].” Fed. R. App. P. 28(a)(8)(A). It makes no difference that the quotation does in fact appear in another case or that the misquoted language is supported generally by Kamalthas. See Shrestha v. Holder, 590 F.3d 1034 (9th Cir. 2010). Rather, that only underscores that it was possible for Sethi to provide the correct citation and that he failed to do so. 14

Sethi's gross misrepresentations of real authorities also violate these rules. For example, Sethi argued in the reply brief that “minor linguistic variations in describing one's role within a political party do not justify an adverse credibility finding.” In support of this contention, Sethi cited a real case, Singh, 491 F.3d at 1096–97, and described the case in a parenthetical as “recognizing that minor translation errors and misunderstandings do not support an adverse credibility finding where the record otherwise supports the claim.” There is no colorable argument that Singh is an “authorit[y]” that supports either of these contentions. Fed. R. App. P. 28(a)(8)(A). Singh concerned the denial of an untimely motion to reopen and equitable tolling. 491 F.3d at 1093–94. It does not discuss adverse credibility or translation errors other than by noting that the petitioner was not found credible and his prior counsel initially failed to submit complete translations of supporting documents. Id. at 1092–93. To the extent that Sethi's argument, or the description of the holding in the parenthetical, was “warranted under existing law,” Sethi failed to identify that law. Cal. R. Pro. Conduct 3.1(a)(2). And his citation to a wholly inapposite case was not a good-faith argument for changing the law. Id.

Had Sethi acted with competence and diligence, he would have read the Lnu briefs before submitting them to this Court, including the authorities cited in the briefs. “Plainly,” however, “counsel did not read the cases he cited before filing his appellate briefs: Had he read them, he would have discovered, as we did, that the cases did not contain the language he purported to quote, did not support the propositions for which they were cited, or did not exist.” Noland, 114 Cal. App. 5th at 445, 336 Cal.Rptr.3d 897.

It is no excuse that Sethi entrusted substantive cite checking to subordinates, and it is no excuse that Sethi purportedly did not know his subordinates had used generative AI. 15 It was Sethi's signature, and his alone, on the briefs. So, it was Sethi who “present[ed]” the unwarranted contentions. Cal. R. Pro. Conduct 3.1(a)(2); Fed. R. Civ. P. 11(b). And Sethi's signature was an attestation that he personally reviewed the contents of the brief, including the cited authorities, and that they were accurate. Rule 32-1 Adv. Comm. Note; see Fed. R. App. P. 32(d). Sethi did no such review, and his attestation was not conditional on the tools that his subordinates might have used to prepare the first draft, nor could it be. See Fed. R. Civ. P. 11(b).

To the contrary, Sethi attests that he and Rounds “do not normally vet citations used by the Brief writer during our review.” That is an extraordinary confession. A competent and diligent attorney cannot decline to “vet” citations, in a brief he signs, for substantive validity—in other words, to read the cited authorities and ensure that they are on point. Cal. R. Pro. Conduct 1.1, 1.3; Cal. Bar Practical Guidance 3 (Duties of Competence and Diligence); Rule 32-1 Adv. Comm. Note.

We do not suggest that every minor typographical error in a citation gives rise to a violation of the ethical and procedural rules. The errors we identify, however, are not plausibly typographical. The two nonexistent cases that Sethi cited bear no reasonable resemblance to real cases. The closest possible cases have different names, reporter numbers, and years of publication. And we have not identified a plausible explanation for how typographical errors might have resulted in the misrepresentations of Singh, 491 F.3d at 1096–97, and Zahedi, 222 F.3d at 1165. Nor has Sethi offered one. While misattributing a quotation can be a typographical error that lawyers might be more familiar with—citations may be misplaced when sentences are rearranged in a brief—Sethi does not contend that the misattributed quotations were typographical errors, so we do not address how our sanctions analysis might have differed with a plausible typographical explanation.

C. Candor

The misconduct in this case did not end with the initial filing of the Lnu briefs. At every subsequent step—including the Motion to Correct, oral argument, the Response to our Order to Show Cause, and more recent filings in other cases—Sethi and Rounds have knowingly or recklessly made false statements to this Court.

Again, under the California Rules of Professional Conduct, “[a] lawyer shall not ․ knowingly make a false statement of fact or law to a tribunal or fail to correct a false statement of material fact or law previously made to the tribunal by the lawyer.” Cal. R. Pro. Conduct 3.3(a)(1). Knowledge “means actual knowledge of the fact in question.” Cal. R. Pro. Conduct 1.0.1(f). “A person's knowledge may be inferred from circumstances.” Cal. R. Pro. Conduct 1.0.1(f); see also In re Domestic P'ship, 118 Cal. App. 5th at 1128, 342 Cal.Rptr.3d 227.

1. Innocent Typographical Errors

Sethi and Rounds each violated their duty of candor when they claimed that the nonexistent cases in the Lnu opening brief were the result of typographical or copy-paste errors. Sethi described the hallucinations as “typographical errors” in the Motion to Correct. At oral argument, Rounds stated that the intended citations were “somewhat garbled” and that “it looks like it was a copy and paste error or something like that.” Oral Argument at 9:05–9:28. And in the Response, Sethi and Rounds continue to insist that at the time of the Motion to Correct and oral argument, they believed the errors were the product of innocent typographical mistakes.

As we have explained, we have not identified any plausible explanation for how innocent typographical or copy-paste errors could turn Udo v. Garland, 32 F.4th 1198 (9th Cir. 2022), and Lai v. Holder, 764 F.3d 1098 (9th Cir. 2014)—the real cases—into “Eduardo v. Garland, 28 F.4th 742 (9th Cir. 2022),” and “Lay v. Holder, 729 F.3d 962 (9th Cir. 2013)”—the clearly hallucinated citations in the opening brief. See, supra, Section III.B.2. Further, Sethi and Rounds's practice of not reading the cases in the Brief Writers’ briefs before filing them created an “inherent risk” of citing fake cases and authorities. In re Domestic P'ship, 118 Cal. App. 5th at 1130, 342 Cal.Rptr.3d 227 (quoting Shayan v. Shakib, 116 Cal. App. 5th 619, 625, 339 Cal.Rptr.3d 354 (2025)). We thus infer from the circumstances that Sethi and Rounds knew as early as the Motion to Correct that the errors were not innocent typographical or copy-paste errors. Sethi and Rounds thus knowingly made a false statement to this Court by claiming otherwise and knowingly made a false statement by denying their prior knowledge in the Response.

We also infer that Sethi and Rounds identified Udo and Lai as the closest possible citations not to correct an innocent error, but to cover up the true source of the errors. Rounds insists on his innocent prior state of mind by defending the substantive validity of the replacement citations. But Rounds only digs himself into a deeper hole. His analysis of Udo is either grossly incompetent or knowingly deceptive. Supra, Section I.C. He either recklessly or deliberately mischaracterizes how the Lnu briefs relied on “Lay” and Lai. Id. The post-hoc rationalizations are thus not forthright. Instead, they add to our inference that Sethi and Rounds have knowingly and repeatedly misled this Court as to the source of the errors and their awareness of the source.

In circumstances such as these, the duty of candor requires a transparent disclosure of the source of the errors. An attorney who erroneously submits a hallucination in a brief must notify the Court and opposing counsel immediately, describe the nature of the error (a fabrication, a gross misrepresentation, etc.), and disclose how the error came about—here, misused generative AI.

Instead of transparency, and despite our Order to Show Cause, Sethi has apparently chosen to engage in more subtle subterfuge. In one of the matters we identified in our Order to Show Cause, Contreras Pelayo v. Bondi, No. 24-5168, Sethi filed a “Notice of Errata” that identified two hallucinated citations and requested to replace them with real citations.  16 No. 24-5168, Dkt. 21. The motion did not represent the errors as “typographical.” But nowhere in that motion did Sethi disclose that the prior citations were hallucinations. Instead, Sethi merely said that the brief “contains errors in two of the case citations.” Id. He then simply listed the fabricated citations and said that each was “an error” or “incorrect,” and then identified what the real citation “should be.” Id.

This is not sufficient disclosure. By citing the hallucinations in the opening brief and signing the brief, Sethi previously attested to the accuracy, and thus the existence of, the hallucinated citations. 9th Cir. R. 32-1; Rule 32-1 Adv. Comm. Note. Swapping a hallucination out for a real case does not correct the prior false claim that “this is a real case.” By failing to notify the Court that Sethi previously cited cases that do not exist, Sethi “fail[ed] to correct false statements of ․ law previously made to the tribunal.” Cal. R. Pro. Conduct 3.3(a)(1). Of course, this Court has the discretion to accept replacement citations in such circumstances. But failing to disclose the nature and source of the errors is conduct unbecoming of a member of this Court's bar. See Fed. R. App. P. 46(b)(1).

2. Denying the Use of Generative AI

Sethi and Rounds also violated their duties of candor by categorically denying the use of AI at oral argument (Rounds) and failing to disclose the use of AI in the Motion to Correct (Sethi). Rounds definitively asserted at oral argument, three times, that generative AI “was not used” to prepare the Lnu briefs. Oral Argument at 9:28–9:32, 10:05–10:15, 10:40–10:44. We infer that Rounds had actual knowledge that this was a false statement because he later admitted as much. Only minutes after definitively and unflinchingly disclaiming any use of AI, and only under continued questioning, Rounds admitted that it was “possible” that the errors could have been the product of generative AI. Id. at 11:06. The nature of the errors we have identified, and the lack of any plausible alternative explanation, further support the inference of actual knowledge. And, again, the Firm attorneys’ practice of not reading cases in draft filings created a high risk that generative AI might have been inappropriately used. See In re Domestic P'ship, 118 Cal. App. 5th at 1130, 342 Cal.Rptr.3d 227. Even if Rounds was not yet definitively aware that generative AI had been used, it is beyond belief that he knew for a fact that it was not used. Candor requires admissions of uncertainty just as much as it requires revealing facts known for certain.

Sethi and Rounds claim that, at the time of the Motion to Correct and oral argument, they did not believe that the errors were the product of generative AI because they had a policy against the use of AI in their firm. The fact of any such policy, which Sethi and Rounds did not attach in written form to their Response, is irrelevant. It does not change the fact that the errors could not plausibly have resulted from innocent typographical or copy-paste errors. And it does not change the fact that Rounds could not have honestly believed that there was no possibility of generative AI misuse at the beginning of oral argument, given his later pivot and the nature of the errors.

We stress that 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. There is no upside to denying the use of generative AI or to passing off an AI hallucination as an innocent typographical error. The other rules discussed in this order—competence, diligence, meritorious arguments, citations to authority, attestations to accuracy—do not turn on the source of the error. Supra, Section III.A.

If, in the Motion to Correct, Sethi and Rounds 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. But that is not what they did. The gravity of discipline we impose, including the temporary suspension of practice, is owed to this repeated failure of candor.

IV. Discipline

For the foregoing reasons, we impose the following discipline under Federal Rule of Appellate Procedure 46(b) and Ninth Circuit Rule 46-2:

1. Sethi and Rounds are each personally sanctioned in the amount of $2,500. Within twenty-one days of the date of this Order, each shall pay these sanctions by check or money order payable to “Clerk, US Courts” and mailed to this Court's Clerk's office in San Francisco. The case number and a copy of this Order should be included with the payment.

2. Sethi and Rounds are hereby suspended from practice before this Court for a period of six months starting ten days after this Order is filed.

3. Sethi and Rounds are ordered to provide a copy of this Order to their clients, opposing counsel, and the presiding judge in every pending state or federal case in which they are counsel of record. They shall also provide a copy of this Order to every attorney in their law firm. They must comply with this requirement within ten days from the date of this Order and must certify to the Court within twenty-four hours of that compliance that the requirement has been met.

4. Sethi, Rounds, and all attorneys at the Firm are ordered to include in all future filings a statement, made under penalty of perjury, addressing whether generative AI was used, disclosing the name of the tool used, and certifying that the attorney signing the brief or other filing has personally reviewed the filing and that all citations and quotations therein refer to existing authority. This requirement shall remain in place for two years from the date of this Order.

5. The Court orders the Clerk of Court to serve a copy of this Order on the State Bar of California and any other applicable licensing authorities for further proceedings as appropriate.

IT IS SO ORDERED.

FOOTNOTES

1.   Sethi Law Group, https://www.sethilawgroup.com/ [https://perma.cc/3JHF-EQVQ] (last visited May 14, 2026).

2.   Only Sethi's name appears on the briefs.

3.   We add quotation marks to the names of fabricated cases to distinguish them from real cases.

4.   Sethi quoted Kamalthas as stating: “An adverse credibility determination is not necessarily a death knell to CAT protection.” And he quoted Avendano-Hernandez as stating: “The BIA must consider all evidence of record, including evidence of country conditions, in assessing the possibility of future torture.”

5.   Sethi also cited “Eduardo” for the standard of review for CAT claims, which Udo does recite, 32 F.4th at 1202.

6.   The opening brief originally cited “Lay” for a proposition regarding “affidavits” specifically rather than “corroboration” generally. Perhaps recognizing that Lai does not mention affidavits, the Motion to Correct asked that the substance of the opening brief be “corrected,” substituting “corroboration” for “affidavits.” Sethi neglected to ask us to correct a second citation to “Lay” following a reference to affidavits later in the brief.

7.   In Lai, the BIA affirmed an adverse credibility determination in part because Lai claimed that his wife was arrested, but he did not produce corroborating evidence from his wife to support that claim. 764 F.3d at 1107. We held that the lack of corroboration did not support the adverse credibility determination, and even if it did, the agency erred in relying on it because it did not put Lai on notice of the need for corroboration. Id. at 1107–08.

8.   The opinion states that “[a] minor inconsistency in identifying the location of a person's persecution, in light of otherwise consistent testimony, cannot form the basis of an adverse credibility finding.” Bandari, 227 F.3d at 1166. The reply brief states: “See Bandari v. INS, 227 F.3d 1160, 1167 (9th Cir. 2000) (holding that an asylum applicant's ‘failure to remember non-material details [about his mistreatment] cannot form the basis of an adverse credibility finding’).”

9.   Rounds did not enter an appearance or sign the briefs in either of these cases. We do not present this as an exhaustive list of all past or present falsities in Sethi and Rounds's briefs. Rather, we limit our discussion and sanctions to the issues identified in the Order to Show Cause.

10.   Sethi and Rounds waived their right to an evidentiary hearing by not requesting one in their response. Fed. R. App. P. 46(c); 9th Cir. R. 46-2(d).

11.   It was Udo who gave the wrong hotel name, not his family. Udo, 32 F.4th at 1201. The family statements did not describe “country conditions,” but corroborated Udo's account of his own persecution. Id. at 1204–05. The agency did not neglect to consider the statements in the CAT analysis because of any mismatch in the hotel name, but simply failed to mention them at all. Id. at 1205.

12.   Courts, commentators, and industry actors have used the term “hallucination” to describe a range of generative AI errors, but the term is “currently being used inconsistently.” Magesh et al., supra, at 221. Popular legal AI tools have apparently adopted a definition of hallucination that only includes fabrications. We agree with Magesh et al. that this is “plainly irrational,” as such a definition “would require us to conclude that a tool that links only to Brown v. Board of Education on every query ․ has provided ‘hallucination-free’ citations.” Id. We thus adopt the broader definition of hallucinations proposed by Magesh et al.

13.   To our knowledge, these are the most recent independent data available. Generative AI is improving rapidly, so it is possible that these hallucination rates have decreased. See Magesh et al., supra, at 231. Until the technology fundamentally changes, however, even advanced legal-specific AI models are “unlikely to fully solve the hallucination problem,” due to, among other things, the complexity of our common law system. Id. at 219–20.

14.   The California Rules of Professional Conduct provide that, as part of an attorney's duty of candor, an attorney may not “knowingly misquote to a tribunal the language of a book, statute, decision, or other authority.” Cal. R. Pro. Conduct 3.3(a)(1)–(2). This duty could be implicated by the unchecked use of generative tools that are known to be unreliable. See Cal. R. Pro. Conduct 1.0.1(f); see also In re Domestic P'ship of Campos & Munoz, 118 Cal. App. 5th 1112, 1128–30, 342 Cal.Rptr.3d 227 (2026).

15.   To the contrary, before the corrective actions noted in the Response, the Firm's system of delegation may have violated California's responsibilities for supervising nonlawyers and its prohibition on the unauthorized practice of law. Cal. R. Pro. Conduct 5.3(a)–(b), 5.5(a). Without reading the cases cited in the drafts, it was not possible for Sethi and Rounds to have “merged” the preparatory work of the Brief Writers into their own final work product. See Crawford v. State Bar of Cal., 54 Cal. 2d 659, 668, 7 Cal.Rptr. 746, 355 P.2d 490 (1960) (quoting Ferris v. Snively, 172 Wash. 167, 177, 19 P.2d 942 (1933)).

16.   As in Lnu, Sethi sought to replace a citation to “Eduardo v. Garland, 28 F.4th 742, 748 (9th Cir. 2022),” with one to Udo v. Garland, 32 F.4th 1198, 1202 (9th Cir. 2022). See, supra, Section I.A.