Direct answer · as of November 18, 2026

In Mata v. Avianca, Judge Castel sanctioned the lawyers who filed non-existent cases generated by ChatGPT. His June 22, 2023 opinion imposed a $5,000 penalty, jointly and severally, on attorneys Peter LoDuca and Steven Schwartz and their firm, Levidow, Levidow & Oberman P.C., payable into the court registry within 14 days. He found bad faith by the two lawyers, citing conscious avoidance and false and misleading statements, and wrote that there is nothing inherently improper about using a reliable AI tool but that Rule 11 makes the lawyer the gatekeeper of accuracy. For how often this has happened since, see the AI hallucination court cases statistics. Sources: Opinion and Order on Sanctions, ECF 54, read November 18, 2026.

Mata v. Avianca is the case that started it all. In June 2023, a federal judge in Manhattan sanctioned two attorneys and their law firm $5,000, jointly and severally, for filing a brief stuffed with six completely fabricated case citations generated by ChatGPT. Browse the later rulings in our AI sanctions tracker. Every attorney using AI for legal research needs to know this case because it's cited in virtually every AI sanctions ruling that followed.

What sanction did the court impose in Mata v. Avianca?

Short answer (last verified October 30, 2026): In Mata v. Avianca (S.D.N.Y. No. 22-cv-1461), Judge Castel's June 22, 2023 Opinion and Order imposed a $5,000 penalty, jointly and severally, on lawyers Peter LoDuca and Steven Schwartz and their firm, Levidow, Levidow & Oberman, for filing fake ChatGPT opinions. It also required letters to the plaintiff and to six named judges.

ItemWhat the order saysSource
Court and caseRoberto Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), U.S. District Court, S.D.N.Y.; Opinion and Order on Sanctions filed June 22, 2023Opinion and Order, ECF 54, p. 1
Who was sanctionedPeter LoDuca, Steven A. Schwartz and the law firm Levidow, Levidow & Oberman P.C., called “Respondents” in the orderOpinion and Order, ECF 54, p. 1
Penalty“A penalty of $5,000 is jointly and severally imposed on Respondents and shall be paid into the Registry of this Court within 14 days”Opinion and Order, ECF 54, p. 34
Other sanctionsWithin 14 days, letters to plaintiff Mata and to each judge falsely named as author of the fake “Varghese”, “Shaboon”, “Petersen”, “Martinez”, “Durden” and “Miller” opinions, then copies of the letters filed with the court. The court declined to order an apologyOpinion and Order, ECF 54, pp. 34, 33
BasisThe court found bad faith on the part of the individual Respondents, based on conscious avoidance and false and misleading statements, and imposed sanctions under Rule 11 or, alternatively, its inherent authorityOpinion and Order, ECF 54, pp. 2, 34
Timeline in the orderThe March 1, 2023 submission first cited the fake cases; the individual Respondents did not begin to come clean until May 25, 2023, after an order to show causeOpinion and Order, ECF 54, p. 2

Last verified October 30, 2026, read in the court's own PDF of ECF 54. The order sanctions the two lawyers and the firm together; the penalty is one $5,000 amount for all three, not $5,000 each. This table does not cover later cases; the AI sanctions tracker lists those.


Background

Roberto Mata sued Avianca Airlines for personal injuries sustained on a flight. His attorneys, Peter LoDuca and Steven Schwartz of the firm Levidow, Levidow & Oberman, were tasked with opposing Avianca's motion to dismiss the case. Schwartz turned to ChatGPT to research case law supporting their opposition brief.

A later matter with the same root problem is in the People v. Crabill case analysis.

The problem wasn't that Schwartz used AI. The problem was what happened next. ChatGPT generated six case citations that looked legitimate on the surface, complete with reporter citations, page numbers, and quotations from the supposed opinions. Schwartz didn't check whether any of the cases existed. He dropped them into the brief, LoDuca signed and filed it, and neither attorney verified a single citation.

When Avianca's counsel couldn't locate the cited authorities, they flagged the issue with the court. Rather than immediately coming clean, Schwartz and LoDuca initially doubled down. Schwartz even asked ChatGPT whether the cases were real, and the chatbot confirmed they were. It wasn't until Judge Kevin Castel demanded the full text of the cited opinions that the truth came out.

Mata v. Avianca, Inc.
No. 22-cv-1461 (S.D.N.Y. 2023)
Court
U.S. District Court, Southern District of New York
Date
2023-06-22
Category
AI Hallucination / Sanctions
Sanctions
$5,000
AI Case Law: Updated October 2026

What Happened

Schwartz's affidavit of May 25, 2023 was the first acknowledgement by any respondent that the cited cases came from ChatGPT. It included screenshots from his phone in which he asked ChatGPT "Is Varghese a real case" and "Are the other cases you provided fake"; ChatGPT answered that the cases were real and could be found in Westlaw, LexisNexis and the Federal Reporter.

Judge Castel held an evidentiary hearing on June 8, 2023. Both attorneys appeared and testified. The court found that six cited cases were entirely fabricated. None existed in any legal database. The fake opinions contained invented judicial reasoning, fabricated quotations, and citations to other nonexistent decisions. Some even attributed rulings to real judges who never wrote them.

The court also noted that the attorneys had been given multiple opportunities to correct the record and failed to act promptly. The doubling-down period between the initial flag and the eventual admission made things significantly worse.


The Ruling

On June 22, 2023, Judge Castel sanctioned LoDuca, Schwartz and their law firm, Levidow, Levidow & Oberman P.C., $5,000 jointly and severally, payable to the court registry. The court found the two attorneys acted with "subjective bad faith" and sanctioned them under Federal Rule of Civil Procedure 11 or, alternatively, its inherent power. The firm was held jointly responsible under Rule 11(c)(1). Read what the order required.

The court held that attorneys have an affirmative duty to verify the accuracy of all legal citations before filing. Using an AI tool doesn't change this obligation. Schwartz's claim that he didn't know ChatGPT could fabricate information wasn't a defense. The order puts it this way: "there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings."

Beyond the fine, the court ordered the respondents to send notification letters to the plaintiff (their own client) and to each judge who was falsely identified as the author of the fabricated opinions. This public accountability component carried reputational consequences that far exceeded the monetary penalty.

Outcome: LoDuca, Schwartz and their firm were sanctioned $5,000 jointly and severally, payable to the court registry. They were also ordered to send notification letters to the plaintiff and each judge falsely identified as the author of the fake opinions.

Mata v. Avianca Sanctions: the 2023 Court Order on ChatGPT's Hallucinated Cases

The sanction comes from one document: the Opinion and Order on Sanctions (ECF 54, 43 pages, PDF), signed by Judge P. Kevin Castel on June 22, 2023 in Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y.). We re-read the order itself on October 24, 2026; every figure below comes from it, not from press coverage.

The order does not ban AI. It says there is "nothing inherently improper about using a reliable artificial intelligence tool for assistance," and then holds the lawyers to the duty to check what they file.

Why This Case Matters

Mata v. Avianca became the most cited AI case in legal history almost overnight. It put every attorney in the country on notice that AI-generated legal research can include completely invented authorities, and that filing those fabrications carries real consequences.

The case triggered a nationwide wave of judicial standing orders requiring AI disclosure and verification. Within months, hundreds of federal and state judges issued rules addressing AI use in court filings. Judge Brantley Starr in the Northern District of Texas issued a May 2023 AI standing order (widely reported as the first from a federal judge) just days before the Mata sanctions came down.

For the full range of consequences since Mata, from fines to bar referrals, see what happens if AI hallucinates in a court filing.

The ruling also exposed a critical gap in legal education and bar requirements. Most attorneys in 2023 had no training on AI tools, no understanding of hallucination risks, and no firm-level policies governing AI use. Mata v. Avianca forced the profession to confront those gaps head-on.


Lessons for Attorneys

The first lesson is obvious but bears repeating: never file a citation you haven't personally verified in an authoritative legal database. Westlaw, Lexis, Fastcase, Google Scholar, or a court's own PACER database. If a case doesn't appear in any of these sources, it doesn't exist. ChatGPT and similar tools generate plausible-looking but fictional citations routinely.

The second lesson is about the cover-up. When opposing counsel flagged the nonexistent cases, the right move was immediate disclosure and a motion to withdraw the brief. Instead, the attorneys asked ChatGPT to verify its own output (which it confirmed) and delayed disclosure. This turned a fixable mistake into a sanctions-worthy offense. Courts treat the failure to promptly correct the record as independent misconduct.

The third lesson is about firm-level responsibility. LoDuca signed the brief but relied entirely on Schwartz's research. Rule 11 makes every signing attorney personally responsible for the accuracy of what they file. If you're signing a brief someone else drafted, you own every citation in it.


The Bottom Line

Mata v. Avianca established the baseline rule for AI in legal practice: the attorney, not the algorithm, is responsible for every word filed with the court. If you use AI for research, verify every citation in a real legal database before it goes into a brief.

AI-Assisted Research. This piece was researched and written with AI assistance, reviewed and edited by Manu Ayala. For deeper takes and the perspective behind the research, follow me on LinkedIn or email me directly.