Mata v. Avianca: The ChatGPT Citation Hallucination That Led to Court Sanctions (2023)

Incident Summary

On June 22, 2023, U.S. District Judge P. Kevin Castel of the Southern District of New York issued a sanctions order against attorneys Steven Schwartz and Peter LoDuca of Levidow, Levidow & Oberman, P.C., and the firm itself, in the matter of Mata v. Avianca, Inc. (No. 22-cv-1461, S.D.N.Y.). The sanction totaled $5,000, payable to the court. The underlying conduct: attorney Schwartz used ChatGPT to research case law for a personal injury motion, and the tool produced six case citations that were entirely fabricated. None of the cases existed. Schwartz submitted the brief without verifying a single citation against an authoritative legal database.

When Avianca's counsel challenged the citations, Schwartz doubled down — submitting a follow-up declaration attesting to the cases' existence and providing fabricated quotes from the invented opinions. The court ordered Schwartz and LoDuca to produce the actual cases. They could not. Judge Castel then convened a hearing and issued the sanctions opinion, which remains one of the most-cited primary sources in discussions of AI hallucination risk in legal practice.

Structured Record

The Six Fabricated Citations

The brief filed by Schwartz cited the following cases as controlling or persuasive authority. None could be located in Westlaw, Lexis, or any other authoritative legal database, because none existed as described.

• Varghese v. China Southern Airlines Co., Ltd., 925 F.3d 1339 (11th Cir. 2019)
• Shaboon v. Egyptair, 2013 WL 12233011 (N.D. Ill. May 13, 2013)
• Petersen v. Iran Air, 905 F.2d 1073 (7th Cir. 1990)
• Miller v. United Airlines, Inc., 174 F.3d 366 (2d Cir. 1999)
• Estate of Durden v. KLM Royal Dutch Airlines, 2001 WL 1142626 (N.D. Tex. Sept. 20, 2001)
• Zicherman v. Korean Air Lines Co., Ltd., 516 U.S. 217 (1996) — this case is real, but the quotations attributed to it in the brief were fabricated

How the Fabrication Was Discovered

Avianca's defense counsel flagged the anomalies in a letter to the court after failing to locate the cited cases. The court ordered Schwartz to produce the decisions. In response, Schwartz submitted a declaration stating — incorrectly — that the cases were real and attaching what appeared to be printed opinions. Those documents were also AI-generated fabrications.

At the subsequent hearing, Schwartz acknowledged he had used ChatGPT and had not verified the output. He stated he was unaware that a large language model could generate plausible-sounding but nonexistent legal citations. LoDuca, the supervising attorney whose name appeared on the brief, had not reviewed the underlying research before signing the filing.

Court's Findings and Reasoning

ChatGPT is a generative artificial intelligence program. It is not a search engine. It does not retrieve existing documents. Instead, it generates text based on patterns in data it has been trained on. It may generate text that appears to be a judicial opinion but is not.

Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), Sanctions Opinion at 10 (S.D.N.Y. June 22, 2023)

Judge Castel's opinion distinguished between good-faith mistakes — which attorneys make — and the conduct at issue, which he characterized as a failure at multiple levels. Schwartz did not verify the citations. LoDuca, as the responsible signing attorney, did not review the research. When challenged, Schwartz compounded the problem by submitting a declaration that falsely attested to the cases' existence.

The court found violations of Federal Rule of Civil Procedure 11(b)(2), which requires attorneys to certify that legal contentions are warranted by existing law. It also found the conduct sanctionable under the court's inherent authority. The $5,000 amount was calibrated to deter, not to be punitive — the court noted the attorneys had no prior disciplinary history and that the case had not caused substantial prejudice to the opposing party beyond the wasted litigation effort.

Professional Responsibility Implications

The sanctions opinion does not explicitly analyze the New York Rules of Professional Conduct in detail, but the conduct implicates at least three rule areas that practitioners should understand.

What the Sanctions Order Did Not Do

Several things are worth noting about the scope of the order. Judge Castel did not refer the attorneys to the New York bar for disciplinary proceedings, though the opinion left open that other bodies might take independent action. The court did not hold that AI-assisted legal research is impermissible — only that unverified AI output cannot be submitted to a court as legal authority.

The $5,000 sanction is modest relative to what Rule 11 allows. The court's restraint appears deliberate: the opinion reads more as a public record establishing the principle than as a punitive measure against these particular attorneys. The detailed factual findings — the transcript of what Schwartz told ChatGPT, the chatbot's responses, the follow-up questions — were published in full, making the opinion unusually useful as a primary source for practitioners and researchers.

Subsequent Developments

Following the sanctions order, multiple federal district courts issued standing orders or local rule amendments requiring attorneys to disclose AI use in filed documents and to certify that AI-generated content has been verified. As of the publication date of this record, courts including the Northern District of Texas, the Eastern District of Texas, and others had adopted or were considering such requirements.

The underlying personal injury case — a claim under the Montreal Convention arising from an alleged in-flight injury — was eventually dismissed on other grounds. The plaintiff's substantive claims were not adjudicated on the merits because of the procedural posture of the case, not because of the sanctions incident.

Primary Source Reference

The full sanctions opinion is publicly available on PACER under docket No. 22-cv-1461 in the Southern District of New York. The opinion is also reproduced in full in several legal databases. The docket number for the sanctions order specifically is ECF No. 54 (June 22, 2023).