Mata v. Avianca, Inc.

The timeline

• Feb 2022Roberto Mata sues Avianca in New York state court over an in-flight injury; the case is removed to the Southern District of New York.
• 1 Mar 2023Mata's counsel files an Affirmation in Opposition to Avianca's motion to dismiss, citing six federal decisions that do not exist.
• 11–12 Apr 2023After Avianca reports it cannot locate the cases, the court orders counsel to produce copies of the decisions.
• 25 Apr 2023Counsel files an affidavit attaching the purported decisions. The “opinions” are also fabrications.
• 4 May 2023The court issues the first order to show cause.
• 25 May 2023Steven Schwartz's affidavit acknowledges ChatGPT as the source. Its exhibits show him asking ChatGPT itself whether the cases are real; the chatbot assures him they are.
• 8 Jun 2023Sanctions hearing.
• 22 Jun 2023Opinion and Order on Sanctions, published at 678 F. Supp. 3d 443.

What the AI did

ChatGPT produced six fully captioned “decisions”, Varghese, Shaboon, Petersen, Martinez, Durden and Miller, with invented docket numbers, Federal Reporter citations, quotations, and internal citations to further nonexistent cases. When Schwartz later pasted the cases back into the chatbot and asked “Is Varghese a real case” and “Are the other cases you provided fake,” ChatGPT answered that it had supplied “real” authorities findable on Westlaw, LexisNexis and the Federal Reporter. Those screenshots became an appendix to the sanctions opinion.

What the court did, and why

Judge Castel sanctioned the two attorneys and their firm under Rule 11(b)(2) and (c) and, alternatively, the court's inherent authority, on a finding of subjective bad faith. The opinion is careful about what it punishes:

“Technological advances are commonplace and 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.”, 678 F. Supp. 3d at 448 (CourtListener docket, Doc 54)

Respondents “abandoned their responsibilities when they submitted non-existent judicial opinions with fake quotes and citations created by the artificial intelligence tool ChatGPT, then continued to stand by the fake opinions after judicial orders called their existence into question.”

The sanction: a $5,000 penalty, jointly and severally on Schwartz, Peter LoDuca and the Levidow firm, paid into the court registry, “sufficient but not more than necessary to advance the goals of specific and general deterrence”, plus letters to the client and to each judge falsely named as the author of a fake opinion. The court declined to order an apology: “a compelled apology is not a sincere apology.”

What it changed

Mata became the citation that other courts cite when they write the rules. Judge Brantley Starr's standing order in the Northern District of Texas (our page: Judge Starr's order) was among the first to require a certification about AI use in filings. The ABA's first formal ethics opinion on generative AI, Formal Opinion 512, followed on 29 July 2024. And the opinion still works as precedent: the Sixth Circuit quoted it in Whiting v. City of Athens (2026), and Judge Manasco counted it in Johnson v. Dunn (2025) among the “modest monetary sanctions” that had failed to deter the pattern, her reason for reaching for disqualification and bar referral instead.

Sources

• Mata v. Avianca, Inc., No. 1:22-cv-01461 (S.D.N.Y.), CourtListener docket; the sanctions opinion is Doc 54, reported at 678 F. Supp. 3d 443.
• Opinion text (PDF mirror, Berkeley Law archive)
• ABA Formal Opinion 512 (29 July 2024)

Related

• The full sanctions record
• Judge Brantley Starr's AI order (N.D. Tex.), one of the standing orders Mata provoked
• Whiting v. City of Athens, the appellate escalation
• Every verified AI standing order

Checked 4 August 2026. Background dates before March 2023 rest on the opinion's own recitation of the record. Information, not legal advice.