Case D-4 · AI Sanctions Case Tracker

Fletcher v. Experian Information Solutions, Inc.

U.S. Court of Appeals for the Fifth Circuit, No. 25-20086, filed February 18, 2026

Response class R4, Dodged It Repeat offender (verified in the record)

The case

Fifth Circuit panel opinion (Judge Jennifer Walker Elrod), filed February 18, 2026, sanctioning attorney Heather Hersh of FCRA Attorneys. The opinion also recounts the court's 2024 AI-subcommittee history and its decision not to adopt a generative-AI certification rule.

What happened

Hersh admittedly used vLex and Thomson Reuters' CoCounsel to draft much of a reply brief in a Fair Credit Reporting Act appeal, then failed to verify the content. The brief contained 16 fabricated quotations (Reuters reports 21 instances of fabricated quotations or serious misrepresentations of law or fact). When the court issued a show-cause order, her response was evasive about AI use. The opinion also records (fn. 10) a prior AI-misuse sanction against Hersh: Powell v. Nelnet (E.D. Tex., June 4, 2025), where the court made a recklessness finding, expressly warned her, and took her sworn declaration of firm-wide AI-compliance measures.

The rule and the sanction

The court imposed a $2,500 monetary sanction, expressly noting that greater candor would likely have produced a lesser penalty, and lamented that the hallucination problem "shows no sign of abating" / "has no end in sight." Existing duties of accuracy and candor were the basis; the court reiterated that "'I used AI' will not be an excuse." Sources: https://www.reuters.com/legal/government/us-appeals-court-orders-lawyer-pay-2500-over-ai-hallucinations-brief-2026-02-18/ ; https://www.nortonrosefulbright.com/en-us/knowledge/publications/792d8bf3/ai-in-litigation-update-on-gen-ai-sanctions-in-2026

How the lawyer responded

R4, Dodged It (the first show-cause response did not mention AI and attributed the errors to "publicly available versions of the cases"; AI use was admitted only after the court's follow-up questions; the court found the named sources not credible and the response "evasive, misleading"; no verification steps were described; the order states lesser sanctions were likely had she been forthcoming; no completed remediation is recorded). [REPEAT ×2, prior episode Powell v. Nelnet (E.D. Tex. 2025), no entry]

How it could have been avoided

Using a legal-specific AI tool (even a reputable one) does not reduce the verification obligation, check every AI-drafted assertion and citation against primary sources. When a court asks about AI use, answer directly and completely; the Fifth Circuit panel made clear that evasion increases the sanction.

Sources