The Lawyer Sanctioned for Citing Court Cases Invented by ChatGPT | inicIA

Direct answer

In June 2023, federal judge P. Kevin Castel fined lawyers Steven A. Schwartz and Peter LoDuca — and their firm, Levidow, Levidow & Oberman P.C. — $5,000 for filing a brief with six court cases that didn't exist: ChatGPT had invented them. The case, Mata v. Avianca, Inc., became the world's most famous warning about the risk of using artificial intelligence without verifying (court order, Mata v. Avianca, Inc., 678 F. Supp. 3d 443, June 22, 2023; Reuters, June 22, 2023).

What happened

Roberto Mata sued Avianca airlines over an injury on a flight from El Salvador to New York: a service cart struck his left knee. The airline moved to dismiss the case. In the opposition brief, attorney Schwartz cited more than half a dozen court decisions — cases like Martinez v. Delta Air Lines or Varghese v. China Southern Airlines — with verbatim quotes and legal analysis that looked completely real.

None of them existed. ChatGPT had fabricated them: case names, citations, excerpts, and reasoning, all invented with total confidence. When opposing counsel tried to locate those decisions and couldn't find them, the judge ordered the lawyers to produce copies. They couldn't, because there was nothing to produce (court order, 2023; Reuters, 2023).

The sanction and the real reason

Judge Castel imposed a $5,000 fine — joint and several among the lawyers and the firm — under Rule 11, with an explicit finding of bad faith: "conscious avoidance" and false and misleading statements to the court.

The detail almost nobody tells is the most important one: the judge did not sanction them for using ChatGPT. He sanctioned them for using it without oversight and for not correcting the error when it was flagged. The court order says that if the lawyers had owned up to the problem immediately, the outcome would have been different; instead, they "doubled down" and took a month and a half to start telling the truth (court order, 2023). Beyond the fine, the judge ordered them to notify their client and the judges whose names had been falsely cited.

The firm told Reuters: "we made a good faith mistake in failing to believe that a piece of technology could be making up cases out of whole cloth" (Reuters, 2023).

September 2026: it happened again

Three years after the Avianca case, the mistake repeated itself. The New Mexico Supreme Court sanctioned attorney Stephen D. Aarons with a $5,000 fine and contempt for filing an appeal — in a murder case — with witnesses and testimony invented by ChatGPT, without verifying the content before signing it. At the August 21 hearing, Justice C. Shannon Bacon questioned how a legal professional could claim ignorance of AI "hallucinations," which are regularly covered in the media (press coverage, September 2026: elperiodico.digital; kchcomunicacion.com; notebookcheck.org).

The lesson repeats because the cause repeats: publishing what AI generates without a human reviewing it.

The lesson for your business: 5 rules for using AI without risk

The lawyer's case is not an argument against artificial intelligence: it's the manual for using it well. These are the rules that separate an AI-powered business from a scandal headline:

  1. AI proposes; the human signs. Nothing carrying your name, your brand, or your money goes out without human review. Final responsibility always belongs to a person.
  2. Verify against the primary source. If the AI cites a fact, a figure, or a case, ask for the source and open it yourself. What has no verifiable source doesn't get published.
  3. Use closed knowledge bases. A well-configured agent answers only with your approved information — your catalog, your prices, your policies — and clearly says when it doesn't know something, instead of inventing.
  4. Define what the agent may promise and what data it may ask for. Written rules: what it offers, what it never promises, what information it requests, and when it hands the conversation to a human.
  5. Correct fast and head-on. If the AI makes a public mistake, own it and fix it immediately. Judge Castel put it in writing: the mistake is forgiven, the cover-up is sanctioned.

Explore the agents cluster

Frequently asked questions

Does ChatGPT invent information on purpose?

No. It generates the most probable text word by word, and sometimes that text is false even though it sounds completely confident. That phenomenon is called hallucination, and it's a known limitation of language models.

So I shouldn't use AI in my business?

Of course you should: with human verification and clear rules, AI multiplies your capacity to serve, sell, and follow up. The lawyer's case shows exactly what happens when the verification step is skipped.

How do I stop my agent from inventing my business's data?

With a closed knowledge base: the agent answers only with your approved information — catalog, prices, policies — and is configured to say it doesn't know something before inventing an answer.

Was the sanction for using ChatGPT?

No. The judge was explicit in the order: the sanction responded to bad faith and to not correcting the error in time, not to the use of the tool. Using AI with oversight and honesty wasn't the problem; publishing without verifying and hiding it was.

Take the next step

A well-configured AI agent answers with your verified information and hands off to a human when it doesn't know something: that's how you get AI's speed without the Avianca-case risk. If you want to deploy agents with clear rules and verification in your business, talk to inicIA on WhatsApp: https://wa.me/17866120930

Chat on WhatsApp

By inicIA