The Lawyers Filed Cases That Their Chatbot Had Invented

1. A Cart and a Deadline

Roberto Mata said a serving cart injured him aboard an Avianca flight. His passenger-injury claim reached federal court in New York, where the airline sought dismissal. The dispute required his lawyers to answer a question about the time limit for bringing the case. Steven Schwartz searched for supporting decisions using ChatGPT.

The search produced cases that appeared to support his argument.

2. Names on the Page

Schwartz supplied research for a filing submitted by his colleague Peter LoDuca. It cited judicial decisions and quoted passages that appeared to answer the airline’s argument. The document gave case names and legal citations, presenting them as existing authorities the court could consult. Avianca’s lawyers began checking the material.

The airline could not find several of the cited decisions.

3. The Request for Copies

Judge P. Kevin Castel asked the passenger’s lawyers to provide copies of the decisions. That request turned the dispute from what those cases supposedly said to whether they existed. The lawyers returned to the chatbot instead of establishing authenticity through the courts or an independent legal database.

The chatbot continued to insist that its earlier cases were real.

4. More Paper, Same Problem

The lawyers submitted supposed copies of the requested opinions. The additional documents contained further anomalies, including names, quotations and legal reasoning the court could not reconcile with genuine decisions. Castel examined the submitted material and the explanations offered for it. Producing longer documents had not supplied the missing judicial records.

The court required the lawyers to explain how the material was created.

5. The Chat History

Schwartz disclosed that ChatGPT had supplied the research and that he had asked it whether the cases were genuine. Its assurances had persuaded him, although they did not establish that any court had issued the opinions. The inquiry also examined LoDuca’s role in signing and submitting the court papers.

The responsibility for the filed documents remained with the lawyers.

6. The Hearing

At a June 2023 hearing, Castel questioned the lawyers about the submissions and their attempts to defend them after doubts had emerged. The resulting order distinguished an initial mistake from continuing to present fabricated material. It addressed their conduct under the rules governing representations made to a federal court.

A written sanctions decision followed on June 22.

7. A Joint Sanction

Castel imposed a $5,000 sanction jointly on the two lawyers and their firm. He also required notifications to their client and to judges whose names had appeared in the fabricated opinions. The order concerned the false authorities and the lawyers’ conduct; it did not resolve the passenger’s injury allegations on their merits.

The passenger lawsuit still required its own final disposition.

8. Two Separate Outcomes

Judge Castel imposed a $5,000 joint sanction on the lawyers and their firm in June 2023 and required notifications concerning the fabricated opinions. The underlying passenger lawsuit was also dismissed, through its own procedural rulings. The sanctions addressed the submitted material and the lawyers’ conduct in that case. They were not a judgment that every legal document involving an AI tool must be treated the same way.

The fabricated citations led to sanctions in the filed case.

Sources

• Original narrative adaptation — CC BY-SA 4.0
• Wikipedia contributors — Mata v. Avianca, Inc. (CC BY-SA 4.0)