AI Regulation Tracker  /  Court ruling

Fifth Circuit Fines Lawyer $2,500 for AI-Fabricated Brief and Not Coming Clean

The U.S. Court of Appeals for the Fifth Circuit sanctioned an attorney for filing a reply brief full of AI-hallucinated quotations, then punished her evasive response to the show-cause order more than the fabrication itself. Every litigator filing in federal court should read the reasoning.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

The U.S. Court of Appeals for the Fifth Circuit has fined a lawyer $2,500 for filing a reply brief riddled with quotations and citations that generative AI invented, and the opinion is notable less for the fabrication than for what the lawyer did after she was caught. In a published decision filed February 18, 2026, in Fletcher v. Experian Information Solutions, No. 25-20086, Chief Judge Jennifer Walker Elrod wrote that counsel "misled, evaded, and violated her duties as an officer of this court" when confronted with the errors.

What the brief contained

The court had issued a show-cause order on December 18, 2025, enumerating 16 fabricated quotations and five additional misrepresentations of law or fact in the reply brief. The invented quotes were attached to real cases, including Fox v. Vice and Cooter & Gell v. Hartmarx Corp., a pattern the court called more insidious than the earlier era of wholly fictional case names because a false quote hangs on a citation that actually checks out. The court reviewed the case summaries on Google Scholar, CourtListener, Justia, and FindLaw and confirmed the quotations appeared in none of them. The only sources that could have produced them, the court found, were the generative AI products Casetext and vLex.

The cover-up drew the penalty

The most instructive part of the ruling is the court's treatment of the response. In her first answer to the show-cause order, the attorney said she had relied on "publicly available versions of the cases" and never mentioned AI at all. Only when pressed with follow-up questions did she admit using it, and even then the court did not credit her claim that she used it merely to "help organize and structure" her arguments. She never explained what steps she took to verify the citations, so the court concluded she took none. The court also rejected her characterization of the AI output as "paraphrased summaries" from "free online case repositories," saying that phrase implied the false quotes came from public sources that could mislead others, when in fact no such summaries existed. Describing an AI-generated sentence that way, the panel wrote, was "evasive, misleading, and sanctionable" in its own right.

Chief Judge Elrod stated the logic plainly: "Had Hersh accepted responsibility and been more forthcoming, it is likely that the court would have imposed lesser sanctions." The fabrication alone was serious, but the evasion is what fixed the amount. For practitioners, that is the operative lesson. A hallucinated citation is a professional failure; concealing it when a court asks is a candor violation that compounds the exposure.

The legal basis

The court grounded the sanction in Rule 46(c) of the Federal Rules of Appellate Procedure, which lets a court of appeals discipline an attorney for "conduct unbecoming a member of the bar or for failure to comply with any court rule," and in the court's inherent power to sanction abuse of the judicial process under Anderson v. Wells Fargo Bank. The panel emphasized that no AI-specific rule was needed. The Fifth Circuit had studied a proposed local certification rule in 2024, took public comment, and declined to adopt it, concluding that existing rules such as Civil Rule 11 and Appellate Rule 46 already require accurate filings.

What it does not do

The decision does not ban generative AI in legal practice. The court acknowledged AI "can be helpful if done properly and carefully" and framed the duty as one of verification, quoting the rule that a lawyer must "ensure that the legal propositions and authority generated are trustworthy." It also does not announce a new standard; it applies old rules to new facts. And while secondary write-ups have described the Fifth as the second federal circuit to sanction AI fabrications after the Ninth, the opinion itself points to a wider trend, citing the Second Circuit's referral in Park v. Kim and the Tenth Circuit's February 9, 2026 decision in Amarsingh v. Frontier Airlines. The better reading is that federal appellate courts are converging, not that the Fifth is merely the second to act.

The court also noted the attorney had been sanctioned before, in Powell v. Nelnet, where a magistrate judge warned that repeat conduct "would likely warrant" monetary sanctions. That history informs the result but does not change the rule for everyone else: check the output, and if a court questions it, say so immediately.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.