A lawyer in the United States has been fined after submitting a murder appeal containing false police testimony and witness statements generated with the help of ChatGPT.
The New Mexico Supreme Court found that attorney Stephen Aarons used ChatGPT while preparing an appeal for his client, Oscar Renee Sandoval, who had been convicted of murder and sentenced to life in prison.
Aarons reportedly gave trial transcripts and other case materials to ChatGPT and used the AI tool to help summarize the proceedings. However, some of the information produced by the chatbot was not present in the original case records.
The court found that the appeal included made-up witness statements and testimony attributed to people who did not exist. It also contained fictional details about the shooter, including descriptions of clothing that were not supported by the trial record.
The problem happened because the AI-generated information was not properly checked before it was included in the official court filing.
Aarons told the court that he did not fully understand how easily an AI chatbot could create information that sounds believable but is actually false. This problem is commonly called an AI “hallucination.”
The New Mexico Supreme Court held Aarons in contempt and ordered him to pay a $5,000 fine. The court also referred the matter to an attorney disciplinary board for further investigation.
The judges were especially concerned because the inaccurate information was submitted during a murder appeal, where mistakes could have serious consequences for a defendant and the legal process.
The case has now been reassigned to a public defender, while Sandoval’s appeal remains pending.
The incident highlights a growing problem with the use of generative AI in professional work. ChatGPT and similar tools can quickly summarize large amounts of information and help users prepare documents, but their answers are not always accurate.
AI systems can sometimes create names, statements, events or other details that were never present in the original information. These errors can be difficult to notice because the generated text may sound confident and realistic.
The legal industry has already faced several cases involving lawyers who submitted AI-generated court documents containing incorrect or nonexistent legal references. This case is more serious because the problems involved invented witness testimony in a criminal appeal rather than only incorrect case citations.
The incident serves as another reminder that AI-generated information should be carefully checked against original documents, especially when it is being used in legal, medical, financial or other high-risk situations.
AI tools can support professionals with research and document preparation, but responsibility for checking the final information still remains with the person using the technology.
