Fletcher v. Experian Info Solutions: Appellate Sanctions for AI-Hallucinated Briefing and Lack of Candor Under FRAP 46(c) and Inherent Authority

Court United States Court of Appeals for the Fifth Circuit
Date February 18, 2026
Case Fletcher v. Experian Information Solutions, Incorporated; Bridgecrest Credit Company, L.L.C., No. 25-20086

1. Introduction

Fletcher v. Experian Info Solutions is an appellate sanctions decision arising from a Fair Credit Reporting Act dispute in which the Fifth Circuit confronted a rapidly recurring litigation risk: the use of generative artificial intelligence (“AI”) to draft briefs that contain fabricated quotations, misleading citations, and inaccurate factual assertions.

The underlying civil case involved Plaintiff Robert Fletcher’s allegation that identity theft led to an auto-finance account being opened in his name. The appeal itself had already addressed sanctions imposed in the district court against trial counsel and his firm, but a separate issue emerged on appeal: the reply brief filed by appellate counsel Heather Hersh contained numerous “hallucinated” quotations and other serious legal and factual misstatements.

The Fifth Circuit issued an order to show cause (December 18, 2025), identifying multiple fabricated quotations and misrepresentations, and requiring counsel to explain “whether and how she verified the accuracy” of her brief. The opinion resolves that show-cause proceeding and imposes a monetary sanction.

Key issues included: (i) whether counsel used generative AI to draft the brief; (ii) whether counsel met her duty to verify citations, quotations, and record assertions; (iii) whether counsel was candid with the court in responding to the show-cause order; and (iv) what sanctioning authority applies when AI misuse corrupts the accuracy of appellate advocacy.

2. Summary of the Opinion

The Fifth Circuit found that counsel used generative AI to draft “a substantial portion—if not all—of her reply brief” and failed to check the brief for accuracy, resulting in repeated misrepresentations of the law and inaccuracies about the record. The court further found that counsel was not forthcoming in her show-cause responses—initially attributing the problems to “publicly available versions of the cases,” later conceding AI use, and never credibly explaining any verification steps.

Invoking Federal Rule of Appellate Procedure 46(c) (discipline for “conduct unbecoming a member of the bar” or violation of court rules) and the court’s inherent power to sanction abuse of the judicial process, the court ordered Heather Hersh to pay $2,500 to the clerk of court within 30 days.

3. Analysis

3.1 Precedents Cited (and How They Shaped the Decision)

The opinion’s precedents do not primarily develop substantive FCRA doctrine; instead, they supply the framework for (i) understanding AI “hallucinations” in legal filings and (ii) grounding sanctions authority and standards in existing disciplinary doctrines.

A. AI “Hallucinations” as an Identified Litigation Problem

  • Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023): The Fifth Circuit uses Mata as the canonical “first high-profile” federal incident of AI-fabricated citations, illustrating how briefs can cite nonexistent cases or misattribute panels. Mata functions here as a cautionary baseline: the risk is well-known and therefore less excusable.
  • Snell v. United Specialty Ins. Co., 102 F.4th 1208 (11th Cir. 2024) (Newsom, J., concurring): Cited to define “hallucinations” in plain terms—AI-generated “facts” that “just aren’t true.” The Fifth Circuit’s reliance reinforces that the concept is now judicially recognized, not merely a tech-industry label.
  • ByoPlanet Int'l, LLC v. Johansson, 792 F. Supp. 3d 1341 (S.D. Fla. 2025): Quoted for the ethical core: to use generative AI “properly and carefully,” counsel must ensure legal authority generated is “trustworthy.” The Fifth Circuit adopts this as a normative benchmark and then finds counsel’s conduct an “abdication” of duty when verification is absent.

B. Existing Sanctions/Discipline Tools Apply to AI Misuse

  • Anderson v. Wells Fargo Bank, N.A., 953 F.3d 311 (5th Cir. 2020): Supplies the Fifth Circuit’s articulation of inherent power—courts may sanction “abuse of the judicial process.” It anchors the opinion’s thesis that no special “AI rule” is necessary to reach this misconduct.
  • Park v. Kim, 91 F.4th 610 (2d Cir. 2024): Used as persuasive authority for the proposition that relying on fake opinions to persuade a court is “an abuse of the adversary system.” The Fifth Circuit parallels its own approach with the Second Circuit’s grievance referral mechanism.
  • In re Violation of Rule 28(d), 635 F.3d 1352 (Fed. Cir. 2011): Cited to confirm that discipline under Federal Rule of Appellate Procedure 46(c) can include monetary sanctions. This is important because FRAP 46 is sometimes viewed as professional discipline rather than fee-shifting; the citation supports the remedy imposed.
  • United States v. Martinez-Martinez, 1999 WL 1330642 (5th Cir. Dec. 15, 1999): Demonstrates that misrepresenting a “critical fact” can trigger FRAP 46(c) show-cause procedures—supporting the Fifth Circuit’s use of show-cause here.
  • Waldon v. Wal-Mart Stores, Inc., Store No. 1655, 943 F.3d 818 (7th Cir. 2019): Persuasive authority defining “conduct unbecoming a member of the bar” to include deliberately misleading the court or “egregious misjudgment,” broadening the conceptual fit for sanctioning AI-driven inaccuracies even absent proof of intentional fabrication.
  • Amarsingh v. Frontier Airlines, Inc., 2026 WL 352016 (10th Cir. Feb. 9, 2026): Noted for recognizing inherent authority in the specific context of briefs containing AI-hallucinated citations, situating the Fifth Circuit’s decision within a growing inter-circuit pattern.
  • Ben E. Keith Co. v. Dining All., Inc., 80 F.4th 695 (5th Cir. 2023): Supports inherent-power sanctions where counsel discovers misrepresentation and fails to correct it; here, the Fifth Circuit extends the logic to counsel’s misleading responses about the source of errors and failure to candidly address AI use.
  • Jenkins v. Methodist Hosp. of Dallas, Inc., 478 F.3d 255 (5th Cir. 2007): Supplies a calibrating principle—when conduct is non-intentional, first-offense, and accompanied by recognition/apology, a mere admonition may suffice. The court distinguishes those mitigating factors as absent here, particularly due to evasion and lack of candor.

C. The Opinion’s Internal Procedural Context

  • Fletcher v. Experian Info. Sols., Inc., No. 25-20086, 2026 WL 37428 (5th Cir. Jan. 6, 2026): The earlier panel decision vacated the district court’s sanctions order against trial counsel due to due-process and statutory constraints (including that 28 U.S.C. § 1927 applies when an attorney “multiplies the proceedings”). The February 18, 2026 opinion is distinct: it addresses appellate counsel’s misconduct in the Fifth Circuit itself and grounds sanctions in FRAP 46(c) and inherent authority.

D. Why Certain Cited Authorities Matter (Even If Not “Holding” Cases)

  • The court’s discussion of its own proposed (but not adopted) AI certification rule (Notice of Proposed Amendment to 5th Cir. R. 32.3) is not precedent, but it frames a key institutional judgment: existing rules (FRCP 11, FRAP 46, inherent authority) already impose an accuracy-and-candor duty. That framing makes the sanction appear less like an innovation and more like a straightforward application of settled professional obligations to new technology.

3.2 Legal Reasoning

A. Findings of Fact: AI Use + No Verification + Lack of Candor

The court’s reasoning begins with credibility and traceability. It catalogues numerous fabricated quotations and five additional “serious misrepresentations of law or fact.” It then tests counsel’s explanations against readily verifiable sources:

  • Counsel claimed reliance on “publicly available versions of the cases.” The court independently reviewed Google Scholar, CourtListener, Justia, and FindLaw summaries and found none contained the quoted language or propositions.
  • The only sources counsel identified that plausibly could generate the errors were “Casetext” (now associated with Thomson Reuters’ CoCounsel) and “vLex,” which the court described as “generative AI products.”
  • Even after direct questioning, counsel did not identify concrete verification steps; the court inferred she “took none.”
  • Counsel’s continued characterization of AI output as “paraphrased summaries,” without producing any such summaries, was found “evasive” and “misleading.”

B. Rule Selection: FRAP 46(c) and Inherent Authority as Sufficient Bases

The Fifth Circuit emphasizes that modern generative AI is “new technology,” but “the same sanctions rules apply.” It then ties the misconduct to two established sources of authority:

  • Federal Rule of Appellate Procedure 46(c): permits discipline for “conduct unbecoming a member of the bar” or failure to comply with court rules, after notice and opportunity to show cause. The court categorizes filing an unchecked, inaccurate brief—replete with fabricated quotations—as “conduct unbecoming.”
  • Inherent power: independently authorizes sanctions for “abuse of the judicial process.” Submitting a brief “riddled with fabricated quotations and assertions” qualifies as such an abuse; misleading the court about the source of errors aggravates the abuse.

C. Culpability and Calibration: Why a Monetary Sanction (Not Just Admonition)

The court treats the sanction as not merely punitive but protective of the judicial process. It relies on factors akin to those described in Jenkins v. Methodist Hosp. of Dallas, Inc.—and explains why leniency is unwarranted:

  • The misconduct was not a single miscitation; it was repeated fabrication/misrepresentation across numerous authorities and record assertions.
  • The court expressly notes that greater candor and acceptance of responsibility likely would have produced “lesser sanctions,” but instead counsel “misled, evaded, and violated her duties as an officer of this court.”

D. The Opinion’s Practical Rule: “Use AI if You Must, But Verify—and Be Candid”

Although the Fifth Circuit does not adopt a new local rule, the opinion effectively articulates an enforceable principle: AI-assisted drafting is not prohibited, but it is ethically compatible with practice only if the attorney verifies accuracy and maintains candor about the source of errors when challenged. Failure on either front can trigger FRAP 46(c) discipline and inherent-power sanctions.

3.3 Impact

A. On Appellate Practice in the Fifth Circuit

  • Verification becomes the centerpiece: The decision signals that the Fifth Circuit will treat unverified AI-generated legal assertions as sanctionable professional misconduct, even without a dedicated “AI certification” rule.
  • Show-cause inquiries will scrutinize toolchains: The opinion models how courts may evaluate explanations—by checking whether “publicly available” repositories actually contain the purported language and by distinguishing ordinary databases from generative products.
  • Candor is an aggravator/mitigator: The court expressly ties sanction severity to counsel’s responsiveness and honesty; evasiveness can independently justify sanctions.

B. On Professional Responsibility and Law-Firm Process

  • Workflow consequences: Firms will likely implement mandatory “citation-and-quotation verification” checklists, require linking citations to PDFs/official reporters, and ban “quote-like” strings unless matched to the underlying opinion.
  • Training and supervision: The opinion’s emphasis on known risks and the growing number of “hallucination cases” suggests that courts may be less tolerant of “I didn’t know” explanations; supervisory lawyers may face heightened expectations to train teams in AI-safe drafting.

C. On the Broader Legal System

  • No need for AI-specific rules (for now): The Fifth Circuit’s institutional choice not to adopt the proposed rule (Notice of Proposed Amendment to 5th Cir. R. 32.3) is reinforced by this opinion’s message: traditional doctrines (FRCP 11-like “stop-and-think,” FRAP 46, inherent authority) can police AI misuse.
  • National convergence: By citing Park v. Kim and Amarsingh v. Frontier Airlines, Inc., the opinion aligns Fifth Circuit practice with a broader appellate trend of treating AI hallucinations as sanctionable abuse of process.

4. Complex Concepts Simplified

“AI Hallucinations”

In this context, an AI “hallucination” is not a metaphorical flourish; it is a concrete litigation risk: the AI generates legal-sounding statements—quotations, case holdings, or record “facts”—that are not actually in the cited sources (or sometimes not in any real source at all).

Federal Rule of Appellate Procedure 46(c)

FRAP 46(c) gives a court of appeals the power to discipline attorneys who practice before it for “conduct unbecoming a member of the bar” or failure to comply with court rules, provided the attorney receives notice and an opportunity to respond (the show-cause process used here). Discipline can include monetary sanctions.

“Inherent Power” Sanctions

Separate from written rules, federal courts have inherent authority to protect the integrity of proceedings. If a lawyer’s conduct abuses the judicial process—such as attempting to persuade the court with fabricated legal support—the court can impose sanctions to deter repetition and preserve the system’s reliability.

Rule 11 vs. 28 U.S.C. § 1927 (Context from the Underlying Case)

  • Federal Rule of Civil Procedure 11 (district court): focuses on whether filings are legally/factually justified after reasonable inquiry; it includes a “safe harbor” process (service before filing).
  • 28 U.S.C. § 1927 (district court): targets lawyers who “multiply the proceedings” unreasonably and vexatiously—generally addressing conduct that prolongs litigation rather than pre-filing defects.

While the February 18, 2026 opinion sanctions appellate counsel under FRAP 46(c) and inherent power, it situates the discussion in a broader sanctions ecosystem and underscores that AI errors can implicate multiple doctrines depending on where and how the misconduct occurs.

5. Conclusion

Fletcher v. Experian Info Solutions crystallizes a practical, enforceable appellate norm: generative AI may assist drafting, but it cannot replace a lawyer’s duty to verify citations, quotations, and record assertions, and it cannot excuse a lack of candor when errors are exposed.

The Fifth Circuit’s $2,500 sanction rests on two durable pillars—FRAP 46(c) discipline for “conduct unbecoming a member of the bar” and inherent authority to sanction abuse of process—confirming that courts need not wait for AI-specific rules to police AI-driven misinformation. The opinion’s broader significance is its message that the legitimacy of adjudication depends on reliable legal sourcing, and that attorneys remain personally accountable for every quotation and proposition placed before the court, regardless of the drafting tool used.