Generative artificial intelligence (Gen AI) tools have become commonplace in legal practice, offering attorneys a means to research, draft and refine legal documents at unprecedented speed. But the promise of Gen AI comes with a well-documented hazard: “hallucinations.” As the Eleventh Circuitexplained, Gen AI “hallucinates” when, “in response to a user's query, it generates facts that, well, just aren’t true—or at least not quite true.”1 Hallucinations in their modern form have grown more insidious than the obvious citation to a wholly invented case; Gen AI now frequently creates fabricated quotations or mischaracterizes a holding from a real case, making errors far more difficult to detect. As of the date of this update, a researcher who maintains a database tracking Gen AI-related court orders in the United States has documented over 1,148cases of hallucinations by lawyers.2 The Fifth Circuit, in one of the decisions discussed in this update, observed that the problem “shows no sign of abating.”3

Courts have found no shortage of tools to address this misconduct. Federal Rule of Appellate Procedure 38, Federal Rule of Civil Procedure 11, 28 U.S.C. § 1927, state rules of professional conduct and courts’ inherent authority have all served as bases for sanctions in recent decisions. Although certain jurisdictions, including New York,4 and many judges have adopted their own rules of procedure governing the use of Gen AI to prepare court filings, no Gen AI-specific rule has proven necessary—indeed, at least one appellate court determined that existing rules were sufficient to deter misconduct related to Gen AI use. Courts have imposed consequences ranging from public admonishment to monetary penalties, removal from a case, denial of fees and referrals to bar authorities.

From recent decisions concerning Gen AI hallucinations, a pattern emerges: how an attorney responds when confronted with Gen AI hallucinations in a filing can be just as consequential as the underlying error itself. Courts have expressly credited candor and cooperation as mitigating factors and have imposed heavier penalties on attorneys who evade, mislead or challenge the legitimacy of the court’s inquiry. At the same time, as the discussion below illustrates, an honest response does not guarantee immunity from serious professional consequences—the gravity of the underlying misconduct remains independently relevant.

With that backdrop, this update analyzes six decisions concerning Gen AI hallucinations issued earlier this year between February and April 2026: (1) Fletcher v. Experian Information Solutions, Inc.;5 (2) In re: Eric Chibueze Nwaubani;6 (3) Whiting v. City of Athens, Tennessee;7 (4) United States v. Farris;8 (5) Gamez v. County of Fresno;9 and (6) Fivehouse v. U.S. Department of Defense.10

Summary of the cases

  • Fletcher v. Experian Information Solutions, Inc.: After counsel submitted a reply brief to the Fifth Circuit containing 16 instances of fabricated quotations and then misled the court about her use of Gen AI, the court imposed a US$2,500 monetary sanction and expressly noted that greater candor would likely have produced a lesser penalty.
  • In re: Eric Chibueze Nwaubani: Counsel’s briefs in an employment discrimination appeal before the Fourth Circuit cited three nonexistent cases. When confronted, the attorney denied using Gen AI and offered what the court found to be attenuated explanations for the errors. The Fourth Circuit's Standing Panel on Attorney Discipline publicly admonished him for conduct that seriously interfered with the administration of justice.
  • Whiting v. City of Athens, Tennessee: Counsel submitted briefs to the Sixth Circuit containing more than two dozen fake citations and misrepresentations of the record. They challenged the court’s order to show cause concerning sanctions as void and improperly motivated and the court imposed the most severe sanctions in this group: reasonable attorneys' fees, double costs, US$15,000 each in punitive sanctions and a disciplinary referral to the Chief Judge of the Sixth Circuit.
  • United States v. Farris: Counsel used a Gen AI platform to draft two briefs that contained fabricated quotations attributed to real cases and misrepresentations of those cases’ holdings. The Sixth Circuit acknowledged the attorney’s candor and his clean 40-year disciplinary record but still found he committed inexcusable transgressions, removed him from the case, denied all Criminal Justice Act compensation for his time on the appeal and referred the matter to multiple judicial and bar authorities.
  • Gamez v. County of Fresno: Counsel submitted a brief containing citations to nonexistent authorities, fabricated quotations and unsupported legal representations. The Eastern District of California accepted his explanation that he believed the AI tool had anti-hallucination protection and accordingly discharged the show-cause order without imposing sanctions—the most favorable outcome among the decisions reviewed here.
  • Fivehouse v. U.S. Department of Defense: An Assistant United States Attorney submitted a brief containing fabricated quotations and misstatements of case holdings drawn from multiple Fourth Circuit opinions. The court found his conduct knowing and intentional and imposed a public reprimand, but limited the sanction in recognition of the severe professional consequences the attorney had already suffered, including the loss of his position.

Fletcher v. Experian Information Solutions, Inc.

On December 18, 2025, the Fifth Circuit issued an order to show cause after the court’s review of counsel’s reply brief revealed 16 instances of fabricated quotations and 5 additional serious misrepresentations of law or fact. The errors involved quotations attributed to real cases that the court concluded either did not contain the quoted language or bore no apparent relation to the propositions for which they were cited. The brief also contained two misstatements of the record, including the assertion that “no formal Rule 11 motion was ever served,” when in fact the motion had been served at least 21 days before its filing.11 Based on the nature and volume of the errors, the court strongly suspected that the attorney had used Gen AI to draft a substantial portion, if not all, of the brief.

Counsel’s response was, in the court's assessment, “evasive” and “misleading.”12 She initially stated that she had “relied on publicly available versions of the cases” and that “certain paraphrased summaries did not match the reporter text,” explanations the court found curious given that the legal databases she named did not contain the fabricated quotations.13 When the court pressed her with a second round of specific questions, she admitted to using Gen AI, but claimed she used it only to “help organize and structure [her] arguments and to break up overly long paragraphs”—a characterization the court found not credible given the breadth and nature of the errors.14 The court concluded that the attorney had used Gen AI to draft the brief but failed to verify its accuracy, compounding the initial misconduct by misleading the court about the source of the errors. Consequently, the court imposed a US$2,500 monetary sanction under Federal Rule of Appellate Procedure 46(c) and the court's inherent authority. In language that has since been cited by other courts, the Fifth Circuit concluded: “Had [the attorney] accepted responsibility and been more forthcoming, it is likely that the court would have imposed lesser sanctions.”15

In re: Eric Chibueze Nwaubani

The Fourth Circuit's Standing Panel on Attorney Discipline initiated disciplinary proceedings against plaintiff’s counsel following its discovery of fabricated citations in briefs he submitted in Bolden v. Baltimore Gas and Electric Co., an employment discrimination appeal. The initial panel identified a citation to a Supreme Court decision that did not exist. The court issued an order to show cause and required counsel to file a revised brief. His reply brief then introduced citations to two additional nonexistent cases. When asked at oral argument where he obtained these cases, the attorney stated he used LexisNexis, Westlaw and Google.

Rather than acknowledging the errors candidly, the attorney then offered what the Fourth Circuit characterized as “attenuated explanations for purported correct citations.”16 He claimed that each fabricated case was a mistake in which he had intended to cite a different, real case. But the court was not convinced: it determined that the replacement cases he identified were largely irrelevant to his arguments and at least one would have actively undermined his client’s position. Despite that, the attorney denied using Gen AI throughout the proceedings. The court found it unnecessary to resolve whether he had used Gen AI, noting that D.C. Rule of Professional Conduct 8.4(d)—prohibiting conduct that “seriously interferes with the administration of justice”—applies to submissions with nonexistent cases regardless of how those errors arose.17 The court found, by clear and convincing evidence, that the attorney’s conduct violated that rule and publicly admonished him. As the panel noted, while a single mistaken citation may not rise to the level of a disciplinary violation, “three nonexistent cases, each with materially different citations than those of the cases [the attorney] meant to cite” plainly did.18

Whiting v. City of Athens, Tennessee

The Sixth Circuit’s March 2026 decision in Whiting addresses what the court described as “over two dozen fake citations and misrepresentations of fact” in briefs submitted by attorneys on behalf of their client in three consolidated appeals.19 The court identified wide-ranging issues: citations to nonexistent cases, quotations that did not appear in the cited opinions, cases cited for propositions they neither discussed nor supported and affirmative misrepresentations of the district court record. Faced with this body of errors, the court issued a detailed order to show cause directing the attorneys to answer seven specific questions: explain why sanctions should not issue; provide Westlaw or LexisNexis copies of every case cited across all three briefs; identify all quoted material; state who wrote each brief; disclose whether the briefs were ghostwritten; confirm whether they used Gen AI and explain their cite-checking process.

The attorneys answered none of the seven questions. Instead, they opposed the court’s order, arguing it was “void on its face for failing to include a signature of an Article III judge,” was “motivated by harassment of the Respondent attorneys,” and “reflect[ed] illegal ex-parte [sic] communications within this Court.”20 The Sixth Circuit rejected the arguments, holding that the attorneys’ briefs—by relying on fabricated case citations and thus misrepresenting the law—rendered their appeal frivolous under Federal Rule of Appellate Procedure 38, irrespective of any legitimate arguments and propositions in the briefs. The court also invoked its inherent authority, finding bad faith in counsels’ refusal to engage with the court’s order and in their use of the court system to pursue a result the applicable law did not permit. The resulting sanctions were the most severe among the decisions covered in this update: reasonable attorneys’ fees, double costs, US$15,000 each in punitive sanctions and referral of the matter to the Chief Judge of the Sixth Circuit for consideration of disciplinary proceedings under Local Rule 46.

United States v. Farris

Farris illustrates an important nuance: candor, while clearly valued by courts, does not guarantee immunity from serious professional consequences when the underlying misconduct is sufficiently grave. In Farris, court-appointed counsel for a defendant under the Criminal Justice Act, having no prior disciplinary history over a 40-year career, directed a staff member to upload district court documents to a Gen AI platform to generate a first draft of his principal brief. It was his first time using the program in that way for an appellate brief. The court immediately identified the brief as AI-generated due to the file name indicating that the brief was drafted by a Gen AI platform.21 On substantive review, the court found that the briefs contained fabricated quotations attributed to two real Sixth Circuit decisions, as well as misrepresentations of both cases’ holdings. For example, the principal brief stated that the Sixth Circuit “reversed role enhancements on similar facts. In Washington,” when in fact Washington upheld such an enhancement.22 The misattributed quotations did not appear in any legal authority the court could locate.

When the court issued its show-cause order, the attorney responded promptly and fully. He admitted to Gen AI use, conceded that the inaccurate quotations were AI-generated and did not appear in any legal authorities, acknowledged that the briefs misrepresented the holdings of the two legitimate Sixth Circuit cases and accepted full responsibility. The Sixth Circuit expressly noted its appreciation for “[the attorney’s] timely response . . . as well as his candor in acknowledging his improper use of artificial intelligence” and took note that he had never before faced misconduct allegations before a court.23 Nevertheless, the court found his transgressions “inexcusable” and imposed consequences that, while not monetary, carried significant professional weight: removal from the case, denial of all Criminal Justice Act compensation for his time on the appeal, referral to the Chief Judge of the Sixth Circuit for consideration of disciplinary proceedings and service of the opinion on the Chief Judge of the Eastern District of Kentucky and the Kentucky Bar Association.24 Farris establishes that candor can influence the form that a sanction takes (the court imposed no monetary fine), but courts retain full authority to impose serious professional consequences even after an attorney responds honestly and cooperatively.

Gamez v. County of Fresno

In Gamez, the Eastern District of California issued an order to show cause after determining that an opposition brief filed by counsel appeared to contain errors stemming from unverified Gen AI use. The brief cited authority that did not exist, fabricated quotations from authority that did exist and made material legal representations without supporting authority. The attorney’s response to the order was candid and specific. He explained that his original draft contained none of the hallucinated authorities and that the errors entered the document when he used a paid legal Gen AI research application to “refine and improve” the draft.25 He chose that particular Gen AI application, he said, because its marketing materials and his prior experience with the program led him to believe it had guardrails that would prevent hallucinations. Had he understood that it could introduce hallucinated authority, he explained, he would have run a final cite-check using non-AI legal research software, as he did for other platforms.26

The court accepted counsel’s explanation in full and discharged the order without imposing any sanctions—the only sanction-free outcome of that kind among the decisions reviewed in this update. In doing so, the court accepted the attorney’s representation that he did not understand that using Gen AI in the format he used could generate inaccurate information, he recognized and apologized for his errors and would ensure future filings would not include these issues. The court still cautioned that Gen AI-related deficiencies in filings are a “serious concern” and emphasized that courts must be able to rely on citations without independently verifying them, warning that counsel would face sanctions if any future deficient filings appeared in the case.27 Gamez is the clearest illustration in these cases of the potential benefit of approaching Gen AI-related errors with transparency and genuine accountability from the outset.

Fivehouse v. U.S. Department of Defense

Fivehouse is the most recent and, in several respects, the most serious decision in this update. A then-Assistant United States Attorney submitted a response brief in an Administrative Procedure Act case that contained fabricated quotations and misstatements of case holdings drawn from multiple Fourth Circuit decisions. For example, the brief cited one Fourth Circuit case for a quotation that, at best, mischaracterized the opinion’s language and then separately attributed three additional propositions to the same case that appeared nowhere in the opinion.28 Opposing counsel detected the fabrications and identified them in a reply brief.

The attorney’s path to disclosing Gen AI use was indirect. His written response to the court’s initial order attributed the errors to an “inadvertent” clerical mistake—specifically, that he had accidentally filed an “unfinalized draft,” making no mention of Gen AI.29 His opening statement at the hearing again omitted any reference to Gen AI. Only after the court began directly questioning him about each individual fabrication did the attorney admit that he had used a Gen AI tool to draft the brief, claiming he had “panicked” after accidentally overwriting his draft and turned to Gen AI to make up for lost time.30 The court found the attorney’s account not credible, noting that he had more than a week remaining under the applicable local rules from the time he claimed to have discovered the problem. The court further concluded, by clear and convincing evidence, that the attorney knowingly and intentionally submitted fabricated materials and identified his failure to be forthcoming as a factor “strongly favoring sanction,” expressly noting that it had considered the Fifth Circuit’s statement in Fletcher, discussed supra, that greater candor would have led to lesser sanctions.31 The court imposed a public reprimand—a sanction “more serious than an admonition”—but limited its order in recognition of the severe professional consequences the attorney had already suffered: he resigned from the US Attorney’s Office the day after the hearing, became the subject of a Department of Justice Office of Professional Responsibility investigation and suffered significant reputational harm.32 The court stated that, absent those consequences, the severity of his misconduct and his position of trust as a federal prosecutor would have warranted significantly more punitive measures.

Several important patterns emerge from these Gen AI hallucination sanctions cases:

  • Existing rules can be used to address Gen AI-related misconduct and courts will not hesitate to apply them.

Federal appellate and civil procedure rules, courts’ inherent authority and state and local bar conduct rules have each served as grounds for sanctions in recent decisions. As the Fifth Circuit’s decision in Fletcher emphasized, “[m]odern generative AI may be a new technology, but the same sanctions rules apply and the rules [courts] have are well equipped to handle these types of cases.”33 Further, all attorneys who sign their names to a brief are responsible for its contents and may face sanctions if the court discovers hallucinations. Attorneys would do well to heed the warning from the Fourth Circuit in Nwaubani—that the ethical prohibition on conduct that seriously interferes with the administration of justice “applies to submitting a brief with nonexistent cases no matter how it is done, whether through generative AI or not.”34

  • An attorney’s candor and cooperation when confronted with Gen AI-related errors may serve as a meaningful mitigating factor.

Courts have explicitly indicated that an attorney’s candor and cooperation when confronted with Gen AI-related errors serves as a meaningful mitigating factor. Fletcher is the clearest statement of this principle: the Fifth Circuit stated outright that a more forthcoming response would likely have produced lesser sanctions. Gamez shows the outcome at the favorable end of the spectrum (no sanctions at all) for an attorney who gave a candid, detailed and apologetic account of how the errors occurred. The Fivehouse court observed that “[c]ourts have shown grace to attorneys who promptly admit their mistake” and are “less forgiving of attorneys who are unwilling or unable to do so.”35

  • Candor is not a complete shield.

These decisions suggest that, while courts may acknowledge an attorney’s transparency and candor in acknowledging Gen AI hallucinations, they may nonetheless impose substantial professional consequences. For instance, the Sixth Circuit in Farris expressly credited counsel’s candor and clean record, yet still removed him from the case, denied all compensation and triggered multiple disciplinary referrals. The takeaway from Farris, read alongside Gamez, is that candor is the right (and ethical) approach but will not necessarily prevent sanctions.

  • Perceived evasiveness and combativeness may significantly aggravate sanctions.

In Whiting, the attorneys refused to answer any of the court’s specific questions and instead challenged the order to show cause, resulting in the harshest overall sanction package, including the only punitive monetary sanctions (US$15,000 each) in this group. Similarly, in Fletcher, counsel gave misleading responses before eventually admitting Gen AI use, which resulted in a monetary fine accompanied by an explicit statement from the court that greater candor would have led to lesser sanctions. And although the attorney from Fivehouse was similarly evasive and only received a public reprimand (a lighter sanction), the court signaled that the sanction would have been more severe but for the consequences the attorney had already faced. Accordingly, attorneys facing possible sanction for Gen AI hallucinations should not only be candid but also clear and non-combative in response to questioning from the court.

  • Courts are signaling that the era of comparatively light sanctions may be drawing to a close.

Documented hallucination cases by lawyers in the United States are rapidly increasing and several courts have noted that warnings, admonitions and public reprimands have not solved the problem. The Fivehouse court stated directly that courts “should begin meeting this challenge with an eye towards deterring similar conduct” and that this will “necessarily and unfortunately, involve moving beyond admonitions and reprimands into more punitive sanctions.”36 Attorneys and law firms should treat the current case law not as a ceiling but as a floor—and plan accordingly.

Practical Takeaways

The six representative decisions analyzed here carry practical implications for any attorney or law firm that uses or considers using Gen AI in legal practice.

Verification is key.

If Gen AI is used to assist with drafting briefs, it is critical to verify each citation and quotation prior to submission. Additionally, attorneys should consider how they delegate this responsibility. The Farris court noted that the attorney of record’s reliance on “staff” rather than on himself or another attorney to supervise Gen AI-generated work product “fell short of his obligations as attorney of record.”37 We do not read this statement to mean that an attorney cannot rely on non-attorney staff to perform this work, but rather that the attorney of record remains ultimately responsible for appropriately supervising their team in accordance with the Model Rules of Professional Conduct (see Model Rules 5.1, 5.3, 5.4 and 5.5 and the comments to these rules).

Understand the Gen AI tools you use, including their limitations.

Gamez shows that a Gen AI tool's marketing assurances about hallucinations cannot be trusted at face value. As Farris observed, attorneys “should not utilize technology without knowing the ways in which it can be misused or contribute to inaccuracies,” which remains true “even when new tools are sponsored by trusted legal technology providers.”38

The pattern across these six cases is sufficiently consistent that attorneys should treat transparency not merely as an ethical obligation but as a mitigation step to minimize the risk of sanctions. At the same time, candor should not generate complacency: as Farris illustrates, an honest response will not necessarily eliminate the risk of sanctions or neutralize the consequences of serious underlying misconduct.

Consider implementing or revisiting firm-wide policies governing Gen AI use.

The Farris opinion notes that relevant steps may include reviewing and validating Gen AI-generated content, implementing firm-wide Gen AI policies, keeping current with jurisdiction-specific guidelines and adhering to ethical billing practices when using Gen AI tools.

Consider jurisdiction-specific guidance.

Courts, bar associations and state disciplinary bodies continue to issue guidance on Gen AI use at a rapid pace. The legal landscape around Gen AI in practice is evolving faster than most traditional areas of law and practitioners who fall behind on that guidance may face exposure.

Consider revisiting outside counsel guidelines.

While these practical takeaways focus on steps lawyers and firms can consider when evaluating their use of Gen AI, it may be useful for companies to consider whether additional instruction or guidance to outside counsel on the use of Gen AI in their matters is appropriate.

Our Team has significant experience counseling clients on the litigation and professional responsibility dimensions of Gen AI use in legal practice. For any questions regarding these issues, please contact us.

Footnotes

  1. 1
    Gamez v. Fresno, No. 1:26-cv-00297-KES-EPG, 2026 WL 925944 (E.D. Cal. Apr. 9, 2026).
  2. 2
    Fivehouse v. U.S. Dep’t of Defense, No. 2:25-CV-00041-M 2026 WL 1278575 (E.D.N.C. Apr. 27, 2026).
  3. 3
    Fletcher v. Experian Info. Sols., Inc., 168 F.4th 231, 237 (5th Cir. 2026).
  4. 4
    Id. at 239.
  5. 5
    Id.
  6. 6
    Id. at 239.
  7. 7
    In re: Eric Chibueze Nwaubani, 2026 WL 687194, at *4 (4th Cir. 2026).
  8. 8
    Id. at *3.
  9. 9
    Id. at *4.
  10. 10
    Whiting v. City of Athens, Tenn., 170 F.4th 455, 458 (6th Cir. 2026).
  11. 11
    Id. at 459.
  12. 12
    USA v. Farris, 171 F.4th 920, 921 (6th Cir. 2026).
  13. 13
    Id.
  14. 14
    Id. at 923.
  15. 15
    Id. at 923–24.
  16. 16
    Gamez v. County of Fresno, 2026 WL 925944, at *2 (E.D. Cal. Apr. 9, 2026).\
  17. 17
    Id.
  18. 18
    Id. at 4.
  19. 19
    Fivehouse v. U.S. Dep’t of Defense, 2026 WL 1278575 at *2–3 (E.D.N.C. Apr. 27, 2026).\
  20. 20
    Id. at 5.
  21. 21
    Id. at 8.
  22. 22
    Id.
  23. 23
    Id. at 9.
  24. 24
    Fletcher v. Experian Info. Sols., Inc., 168 F.4th 231, 239 (5th Cir. 2026).
  25. 25
    In re: Eric Chibueze Nwaubani, 2026 WL 687194, at *4 (4th Cir. Mar. 11, 2026).
  26. 26
    Fivehouse v. U.S. Dep’t of Defense, 2026 WL 1278575, at *8 (E.D.N.C. Apr. 27, 2026).
  27. 27
    Id. at 10.
  28. 28
    USA v. Farris, 171 F.4th 920, 923 (6th Cir. 2026).
  29. 29
    Id. at 922.

Contact