The Third Circuit's first AI-hallucination sanction: a public reprimand for unverified, AI-generated citations

Verification status: PENDING — checker-graduated at confidence 1.000; awaiting the owner’s human sign-off. Not verified (green is human-only).

The problem

Writing for the panel, Judge Chung opened: “An attorney submitted a brief to this Court which contained legal authority summarized and provided to him by a non-attorney. The attorney failed to verify the legal authority in any way. Unfortunately, the research contained myriad inaccuracies” (source [1]). The attorney — Daniel A. Pallen, Esq., admitted to the Pennsylvania and Third Circuit bars — represented the petitioner in McCarthy v. United States Drug Enforcement Administration, No. 24-2704 (source [1]).

What was built

To argue the DEA had acted inconsistently with prior practice, Pallen “provided summaries of eight DEA adjudications” (source [1]). According to the opinion, “in including and preparing these summaries, Attorney relied on case overviews generated by AI and provided to him by a non-attorney,” and “modified some of the [provided] language for ease of reading and then simply incorporated the same into the brief” without verifying the citations (source [1]). The court found: “The summaries for seven of the authorities were riddled with factual and legal inaccuracies, and one of the authorities simply did not exist” (source [1]). After the court ordered him to produce copies of the cited authorities, “Attorney disclosed, for the first time, that the seven summaries were inaccurate and that one authority was hallucinated by AI” (source [1]). The Legal Intelligencer, reporting the case, described “summaries of eight Drug Enforcement Administration adjudications that were generated by AI,” of which “seven … were filed with factual and legal errors, and one … involved a case that did not even exist” (source [2]).

The outcome

“Considering the above, the Court will impose the sanction of a reprimand” (source [1]). Under the Circuit’s Disciplinary Rules, “the Clerk will notify … all other courts before whom [Attorney] is admitted to practice and the National Disciplinary Data Bank” of the discipline and provide a copy of the opinion (source [1]). No monetary sanction was imposed: as a Stevens & Lee client alert put it, the court “chose to reprimand the lawyer” (source [3]), and The Legal Intelligencer noted “the court was electing not to issue monetary sanctions” (source [2]). Because this was the Third Circuit’s first occasion to address attorney AI misuse, the court treated the absence of prior guidance as mitigating: “Had this not been our first opportunity to address the use of AI … it is likely that monetary sanctions would have been imposed here. As this precedent has now been set, the first mitigating factor will not apply in the future and violators may well face any of the sanctions available per Circuit Disciplinary Rules 4.1-4.2” — which include suspension and disbarment (source [1]). Judge Roth, concurring in part and dissenting in part, “would have imposed sanctions more severe than the” reprimand (source [1]).

A note on sourcing

The entire account rests firsthand on the court’s own precedential opinion (Source 1, Tier 1) — every critical quote above is drawn from it. The two press sources (Tier 2) corroborate but do not carry the weight alone. The weakest load-bearing item is the archival status, not any single claim: archive.org’s Wayback Machine returned global 503/502 errors during verification, so the two secondary reports are not yet pinned to a Wayback snapshot; the primary, however, is a durable federal-government (GovInfo) PDF with a recorded SHA-1 and is independently re-fetchable, so the evidentiary anchor does not depend on Wayback.

How this was verified

• Method: Every quoted figure and finding was read verbatim from the fetched sources this session — the Third Circuit opinion (GovInfo PDF, SHA-1 d66e8e801d036c4f44e288560c28f3cf31714dcf) plus the two legal-press secondaries, each re-checked against a local capture. No number is paraphrased, rounded, or recalled.
• Standard: Court-origin. The outcome is the adjudicated discipline, so the record — not any subject’s confirmation — is the proof; green never depends on the sanctioned attorney or the court agreeing.
• Independence: Two mutually independent Tier-2 secondaries (The Legal Intelligencer via TortTalk; a Stevens & Lee alert co-authored by a retired Third Circuit judge) each carry the reprimand and the AI-hallucination finding firsthand; neither relays the other.
• Open item: Binding Wayback snapshots of the two secondaries once archive.org is reachable again. The primary anchor is already durable.
• Verified on: 2026-08-17 (presentation rebuilt 2026-08-18).

Related case files

• LNU v. Blanche (9th Cir.) — attorneys Sethi and Rounds fined $2,500 each and suspended six months for AI-fabricated citations — the same “AI-fabricated authorities in an appellate brief” pattern, but with a monetary sanction and suspension where McCarthy drew only a reprimand.
• In re Nwaubani (4th Cir.) — public admonishment of an attorney for AI-fabricated citations before a standing panel — a sister circuit reaching a comparable non-monetary discipline for the same conduct.
• Noland v. Land of the Free (Cal. App.) — first published California appellate opinion sanctioning AI-fabricated citations, $10,000 — another appellate court treating its case as a matter of first impression.
• Kruse v. Karlen — pro se appellant sanctioned $10,000 after 22 of 24 citations were AI-fabricated — the fabrication-rate extreme against which the Third Circuit’s eight-summary record can be read.

Sources

1. U.S. Court of Appeals for the Third Circuit · McCarthy v. United States Drug Enforcement Administration, No. 24-2704 — precedential opinion (public reprimand of Daniel A. Pallen) · filed 2026-03-27 · https://www.govinfo.gov/content/pkg/USCOURTS-ca3-24-02704/pdf/USCOURTS-ca3-24-02704-1.pdf — Tier 1 (primary, court-origin durable federal PDF; SHA-1 recorded).
2. The Legal Intelligencer (Riley Brennan), via TortTalk · AI Hallucinations Appear in Third Circuit Brief · 2026-03-30 · http://www.torttalk.com/2026/04/ai-hallucinations-appear-in-third.html — Tier 2 (independent reputable legal press naming the parties).
3. Stevens & Lee (ret. 3d Cir. Judge Thomas I. Vanaskie & Karl S. Myers) · AI Hallucination Reaches the Third Circuit · 2026-05-06 · https://www.stevenslee.com/appellate/ai-hallucination-reaches-the-third-circuit/ — Tier 2 (independent firm client alert co-authored by a retired Third Circuit judge).