Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y. 2023), is the case most people mean when they say "the ChatGPT lawyer case." Attorneys representing the plaintiff submitted a brief opposing a motion to dismiss that cited several cases, including purported airline-liability precedent, that opposing counsel and the court could not locate. When pressed, it emerged that the research had been generated using ChatGPT, which had produced citations, including case names, docket numbers, and quoted excerpts, for cases that did not exist. The court, presided over by Judge P. Kevin Castel, ultimately imposed sanctions.
Why this case is still the reference point
It was not the first instance of a fabricated citation reaching a court, but it was the first to receive widespread mainstream press coverage, which is why it functions as shorthand for the entire category of failure. The order in that case is also unusually detailed about the court's own process for confirming the citations did not exist, which makes it useful reading for understanding exactly how these things get caught. See our detailed breakdown of what the sanctions order actually says.
The lesson most retellings miss
The common takeaway is "don't use ChatGPT for legal research." The more useful takeaway, more than two years and (per the Charlotin public database) well over 1,700 subsequent documented cases later, is that the failure was not really about which tool was used. It was about verification: nobody in that filing chain checked the citations against a primary source before filing. The same failure pattern recurs across firms of every size and every category of AI tool, including purpose-built legal AI products, per the Stanford RegLab/HAI benchmark study.
What changed and what didn't
What changed: courts are now less likely to treat a first-time AI-hallucination incident as a novel, sympathetic mistake, and more likely to apply existing Rule 11-style sanctions frameworks directly, since the risk is now well known. Cases like Park v. Kim, 91 F.4th 610 (2d Cir. 2024), Wadsworth v. Walmart Inc. (D. Wyo. 2025), and Coomer v. Lindell (D. Colo.) show courts applying sanctions with less patience for "I didn't know AI could do this" as a defense. What did not change: the underlying incentive to use AI drafting tools under deadline pressure, and the absence, in most practices, of a fast, repeatable verification step before filing.
Expert perspective
Legal commentators tracking the Charlotin database have observed that the rate of new documented cases has not meaningfully declined since Mata, despite the widespread press coverage. That is the clearest evidence that awareness alone does not fix the problem — a mandatory, low-friction verification step does.
The actual fix
Verify every citation against a primary source before filing: confirm it exists, confirm the quotation matches the opinion, and confirm the case actually supports the proposition it's cited for. See our step-by-step verification guide for the full process.
A common question
Were the Mata attorneys the only ones sanctioned in that case?
No — the sanctions order addressed multiple attorneys involved in the filing chain, reflecting the court's view that the certification duty under Rule 11 extends to everyone who signed or supervised the filing, not just whoever ran the original AI query.
Related reading
- Mata v. Avianca: What the Sanctions Order Actually Says
- Rule 11 Sanctions for AI Hallucinations: The 2026 Case Roundup
- Park v. Kim: What an Appellate AI Citation Mistake Costs
- Wadsworth v. Walmart: Inside the Morgan & Morgan AI Sanctions Case
- How to Check for AI Hallucinations in Legal Briefs
Check a brief before you file it →
What the sanctions order specifically required
Beyond the monetary sanction, the court in Mata required the attorneys to send a copy of the sanctions order to every judge falsely identified as the author of one of the fabricated opinions, a remedy specifically designed to correct the record for judges whose names had been attached, without their knowledge or consent, to invented judicial writing. This remedy has been echoed in several subsequent documented cases, reflecting judicial concern not just about the immediate litigants but about the integrity of the broader public record.
Why this case keeps resurfacing in later opinions
Judges evaluating subsequent AI-hallucination incidents frequently cite Mata directly in their own orders, both as explanatory shorthand for readers unfamiliar with the underlying technology and as evidence that the risk was clearly foreseeable by the time of any later incident.
The human details the case is remembered for
Beyond the legal doctrine, the record includes a detail that made the case resonate widely: when the attorney asked ChatGPT directly whether the cases were real, the tool reaffirmed that they were, even offering to provide further details about the fabricated opinions. This exchange, entered into the court record, became a widely cited illustration of why asking an AI tool to verify its own prior output is not a reliable check.
Final takeaway
Mata remains worth reading in full, not just as a cautionary headline, because the order itself lays out in granular detail exactly how the fabrication was investigated and confirmed.
A common question
Was ChatGPT itself ever a party or subject to sanctions in the case?
No. Sanctions in Mata, as in every documented AI-hallucination case, fell on the human attorneys who signed and filed the brief, not on the AI vendor. Courts have consistently located responsibility with the professional obligated to verify, not the tool used to draft.
Related reading
- Mata v. Avianca: What the Sanctions Order Actually Says
- Rule 11 Sanctions for AI Hallucinations: The 2026 Case Roundup
Two years on, Mata is not a relic of an early, less-informed moment in AI adoption; it is the founding entry in a public record that has only grown since, and every attorney using AI tools today is operating with full knowledge of exactly what happened when verification is skipped.
Share this case study with anyone at your firm who still treats AI hallucination as a hypothetical, distant risk rather than a documented, recurring pattern.
That single fact is worth remembering the next time a filing deadline makes verification feel optional.