The court cases that never existed

Roberto Mata sued the airline Avianca, saying a serving cart hurt his knee on a flight. In March 2023 his lawyers filed a brief citing earlier court decisions to support his case.

Avianca's lawyers could not find several of those decisions. Neither could the judge. One of Mata's lawyers had used ChatGPT for research. The cases, quotes and citations it supplied were invented.

The lawyer later said he had asked ChatGPT whether one case was real, and it said yes. On 22 June 2023, Judge P. Kevin Castel fined the two lawyers and their firm US$5,000. They also had to write to each real judge falsely named as an author of a fake ruling.

A chatbot like ChatGPT writes by predicting what text usually comes next. Legal citations follow a very regular pattern: names, a volume number, a court, a year. The model can produce text in that pattern without any real case behind it.

Asking the same chatbot "is this real?" does not help. It is the same pattern-machine checking its own guess. The judge noted that using AI is not wrong in itself. The failure was not checking the output in a real legal database, and then standing by it after questions were raised.

The idea behind this caseLesson: Check what AI tells you →

1. 1Never cite a source you have not opened and read yourself.
2. 2Check every case name, law or statistic from a chatbot in an official database or the original document.
3. 3Do not ask the chatbot to confirm its own facts. Use an independent source.
4. 4If someone questions your sources, re-check them at once and correct the record quickly.
5. 5Agree team rules on which AI tools may be used for research, and require a human check before anything is filed or published.