The canonical AI-hallucination sanction: $5,000, jointly and severally, for the lawyers who filed six ChatGPT-fabricated cases in Mata v. Avianca

The problem

Generative AI produces legal citations that look exactly like real ones — reporter volumes, docket numbers, judges’ names — whether or not the cases exist. In March 2023 that failure mode reached a federal courtroom for the first time at scale: a New York personal-injury lawyer, researching a Montreal Convention limitations question, filed an opposition brief built on cases ChatGPT had invented. The court’s own account is blunt: the Respondents “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” (source).

This is a verified negative: an independent federal court, in a signed public opinion, measured what that workflow produced and priced it. It is the case every later AI-sanction ruling cites — including the Coomer v. Lindell order already in this corpus (source).

What the court found

Six decisions cited in the March 1, 2023 Affirmation in Opposition — “Varghese”, “Miller”, “Petersen”, “Shaboon”, “Martinez” and “Durden” — were, as Respondents themselves ultimately acknowledged, “generated by ChatGPT and do not exist” (source). Two independent newsrooms carried the same count the same day: the brief “included six fictitious case citations generated by an artificial intelligence chatbot” (source), and AP reported the court acted “when it was noticed that six legal cases listed to support their March 1 written arguments did not exist” (source).

The court dissected the flagship fake, “Varghese v. China Southern Airlines”: its “legal analysis is gibberish”, the opinion showing “stylistic and reasoning flaws that do not generally appear in decisions issued by United States Courts of Appeals” (source).

The colloquy from the June 8 hearing is the record’s core artifact (source):

THE COURT: And what did you find when you went to look up the full Varghese decision? MR. SCHWARTZ: I couldn’t find it. THE COURT: And yet you cited it in the brief to me.

What turned an embarrassing mistake into sanctions was the two months after Avianca’s lawyers flagged the citations: instead of withdrawing the brief, counsel stood by the fake opinions, and the court found “bad faith on the part of the individual Respondents based upon acts of conscious avoidance and false and misleading statements to the Court” (source).

The order

On June 22, 2023, the court sanctioned Schwartz, LoDuca and the Levidow Firm under Rule 11: “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 of this Opinion and Order” — a single joint penalty the court called “sufficient but not more than necessary to advance the goals of specific and general deterrence” (source). Reuters reported the same single figure — “a $5,000 fine in total” (source).

The non-monetary sanctions were the sharper edge: the court ordered counsel to “send via first-class mail a letter individually addressed to each judge falsely identified as the author” of a fake opinion (source); CNBC reported that the judge “ordered them to notify each judge falsely identified as the author of the bogus case rulings about the sanction” (source). No apology was ordered — “a compelled apology is not a sincere apology” (source).

Eight days later the court’s own administrative order directed the Clerk “to transfer the penalty in the amount of $5,000 from the Registry of this Court to the Miscellaneous Fines, Penalties, and Forfeitures Fund” — the docket’s own confirmation the penalty was received (source).

The same day as the sanctions opinion, in a separate order, the court granted Avianca’s motion to dismiss the underlying claim as time-barred: the client lost the case too (source).

Why this matters for AI-led operations

Every critical figure in this story originates with an independent measurer — a federal judge with no stake in any AI vendor and no marketing narrative to serve. The court itself drew the operational line that TIN’s verification thesis rests on: “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” (source). An AI step inserted into a professional workflow with no verification step behind it produced six fabricated authorities in one brief; the cost — monetary, and reputational by court-ordered disclosure — was borne by the professionals who signed the output.

A note on source strength. The three critical figures — the $5,000 joint penalty, the six ChatGPT-fabricated cases, and the order to notify each falsely-named judge — each rest on the Tier-1 court PDFs (Docs. 54 and 56), byte-tied to Wayback captures, and are independently corroborated by AP, Reuters and CNBC. The weakest load-bearing source is the dismissal detail: it rests on source, which was fetched live from RECAP but has no Wayback snapshot yet (archival gap flagged); it supports context only, and the CNBC relay corroborates it.

How this was verified

• Method: All three court PDFs (Doc. 54 sanctions opinion and Doc. 55 dismissal opinion of 2023-06-22; Doc. 56 penalty-transfer order of 2023-06-30) fetched live from the public RECAP archive; Docs. 54 and 56 byte-tied (SHA-1 = Wayback CDX digest) to Wayback captures; AP, Reuters and CNBC day-of relays captured; every quote verified mechanically against stored captures.
• Date verified: 2026-08-03.
• Not claimed: “$5,000 each” (a press misstatement — the penalty is one $5,000, jointly and severally); “first ever” framing; the Second Circuit appeal’s outcome (artifact not fetched); the ChatGPT chat-log exhibits (cited only as quoted inside the opinion).

Sources

1. U.S. District Court, S.D.N.Y. — Opinion and Order on Sanctions, Mata v. Avianca, Inc., No. 22-cv-1461 (PKC) [Doc. 54] · 2023-06-22 · https://storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.nysd.575368.54.0.pdf — Tier 1 (court-signed public opinion; origin document).
2. U.S. District Court, S.D.N.Y. — Opinion and Order granting the motion to dismiss (time-barred) [Doc. 55] · 2023-06-22 · https://storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.nysd.575368.55.0.pdf — Tier 1 (court order; archival gap flagged, context only).
3. U.S. District Court, S.D.N.Y. — Order transferring the $5,000 penalty to the Fines fund [Doc. 56] · 2023-06-30 · https://storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.nysd.575368.56.0.pdf — Tier 1 (court order; penalty-received confirmation).
4. Associated Press — “Lawyers submitted bogus case law created by ChatGPT. A judge fined them $5,000” · 2023-06-22 · https://apnews.com/article/artificial-intelligence-chatgpt-fake-case-lawyers-d6ae9fa79d0542db9e1455397aef381c — Tier 2 (independent national newsroom).
5. Reuters (Sara Merken) — “New York lawyers sanctioned for using fake ChatGPT cases in legal brief” · 2023-06-22 · https://www.reuters.com/legal/new-york-lawyers-sanctioned-using-fake-chatgpt-cases-legal-brief-2023-06-22/ — Tier 2 (independent national newsroom).
6. CNBC — “Judge sanctions lawyers for brief written by A.I. with fake citations” · 2023-06-22 · https://www.cnbc.com/2023/06/22/judge-sanctions-lawyers-whose-ai-written-filing-contained-fake-citations.html — Tier 2 (independent national newsroom).

Related case files

• Coomer v. Lindell (MyPillow) — the direct descendant that cites Mata, with 30 defective AI citations and $3,000 sanctions each.
• Wadsworth v. Walmart — the same $5,000 Rule 11 figure applied to a large-firm brief.
• Gauthier v. Goodyear — a smaller sanction that adds a generative-AI CLE requirement.
• McCarthy v. DEA — a federal appellate first-impression reprimand in the same lineage.