Court sanctions lawyers after fabricated ChatGPT citations

On June 22, 2023, Judge P. Kevin Castel imposed a joint $5,000 penalty and notification requirements on two attorneys and their firm in Mata v. Avianca. The chronology dates sanctions, not the March 1 submission.^[1]

What the court found

The court found that nonexistent ChatGPT-generated opinions were submitted and defended after warnings. It found bad faith through conscious avoidance and false or misleading statements. It also distinguished reliable AI assistance from attorneys’ responsibility to check filings (pp.1–2).^[1]

The participant’s account

The order quotes Schwartz’s hearing testimony: “I just never thought it could be made up.” (Finding 11, p.6; hearing transcript p.33.) This is testimony reproduced in the court’s opinion, not a separately audited full hearing transcript.^[1]

From fluent text to evidence

Imagine a library label attached to an empty shelf. The label can look convincing while failing to identify a book you can inspect. This original analogy illustrates the difference between a formatted citation and evidence of an actual source.

Read Hallucination and the survey for vocabulary, then the uncertainty research for a different question: how meanings vary across generated answers. These papers do not reconstruct the case or validate a legal-workflow remedy. The incidents guide compares the evidence with other failures.

Review scope: selected original findings, introduction and conclusion checked. Full underlying docket, complete hearing and subsequent compliance were not independently audited.