ChatGPT-using lawyer punished for citing fake testimony from made-up witnesses

ChatGPT-using lawyer punished for citing fake testimony from made-up witnesses

As an Amazon Associate I earn from qualifying purchases.

Woodworking Plans Banner



“You buried your head in the sand ”

“I didn’t understand that AI might hallucinate truths,”New Mexico defense attorney states.

Credit: Getty Images|NurPhoto

The New Mexico Supreme Court held a ChatGPT-using attorney in direct contempt of court for sending a quick with “incorrect testament from entirely produced witnesses,” consisting of phony authorities statement and other errors. The state’s leading court referred the legal representative to a disciplinary board for more procedures and concluded that he “showed an absence of regret and an absence of issue for his customer.”

Lawyer Stephen Aarons “confessed to the Court that he did not confirm the accurate claims and legal authority in his AI-generated quick before signing it and submitting it with the Court, which he did not notify his customer of this failure or that the quick in chief included numerous accurate and legal misstatements,”the state Supreme Court stated in an order on Wednesday.

Aarons has actually been a criminal defense attorney in New Mexico for over 40 years and was employed by an accused’s member of the family to appeal a murder conviction. Aaron’s now-former customer, Oscar Renee Sandoval, was sentenced to life in jail in February 2025 after being founded guilty of eliminating Shiereen Al-Jibury, who was his partner and the mom of his kids.

A little over a year earlier, in August 2025, Aarons sent the short consisting of phony testament and other mistakes. Weeks later on, the state of New Mexico submitted a movement to strike parts of that quick.

“Respondent confessed to the Court that the short in chief included incorrect statement from entirely made witnesses– Officer Michelle Amarillo, Officer Sanchez, Manal Al-Jibury, and Teresa Marquez,” the court order on Wednesday stated.

Legal representative fed trial records into ChatGPT

Aarons even more confessed sending “incorrect testament from Danny Stanton that he got risks,” “incorrect statement from Linda Stanton about the hazards her spouse got,” and “incorrect statement from Mariah Chavez and Teresa Marquez (produced witness) concerning the shooter’s clothes and look,” according to the order, which stated Aarons likewise “misrepresented legal authority” in citations to previous cases.

Numerous attorneys have actually been captured pointing out phony cases in briefs or erroneously explaining genuine cases. While Aarons didn’t point out phony cases, he erroneously explained genuine ones and pointed out phony testament.

Aarons informed the state Supreme Court at a hearing on August 21 that he fed a computer-generated records of the murder trial and other files associated with the case into ChatGPT, which outputted the phony quotes.

“It’s of little convenience to understand that my stupidity is what brings us together this afternoon,” Aarons informed the court. Aarons confessed his short priced estimate “a number of witnesses who were never ever called at trial,” another “witness who was called however the quick got the name incorrect,” which his short incorrectly explained precedents.

Aarons suggested that he utilized a variation of ChatGPT powered by the OpenAI o3 design, which was launched previously in 2025. “I presumed that it created a bulletproof summary of procedures,” he stated at the hearing, discussing that he believed it would be precise due to the fact that of how prevalent AI usage remains in the legal and medical fields.

Aarons was disallowed from appearing before the New Mexico Supreme Court pending the result of any disciplinary board examination and procedures. He was fined $5,000, to be paid to the State Bar of New Mexico Client Protection Fund, however more charges might come out of the disciplinary procedures bought by the court.

The state Supreme Court bought the general public protector workplace to select a brand-new legal representative for the accused, struck all previous briefs from the record, and stated the case will continue in the court’s 2026-27 term.

Justice: “You buried your head in the sand”

Justices berated Aarons throughout last month’s hearing. They revealed surprise that Aarons didn’t understand AI tools might create incorrect info and mentioned that lawyers need to confirm the precision of details no matter where it originates from. Whether an attorney gets assistance from an AI tool, a law trainee, or a fellow lawyer, the attorney signing the quick should vouch for its precision, they stated.

Justice C. Shannon Bacon was especially withering in her criticism of Aarons. She informed Aarons that there are “a minimum of 8 or 9 arrangements in the standard procedure that you broke.” Bacon stated she was “actually fighting with” Aarons’ claim to be uninformed of AI hallucinations, keeping in mind that her 13-year-old nephew and 75-year-old stepmother are both familiar with the issue.

“So counsel, do you enjoy the news? Do you listen to the radio? Do you check out anything about what’s going on the planet? Since the issue with legal representatives counting on AI hallucinations is an above-the-fold story each and every single day,” she stated. “So either you buried your head in the sand– which’s an option to do that, a deliberate option to be uninformed– or you took a gamble, and neither of those follow the standard procedure.”

Aarons reacted that he sent the quick a year back and “a lot has actually come out in the in 2015.” The issue of attorneys utilizing AI in briefs that mention fabricated cases has actually been in the news frequently for well over 3 years.

Aarons included that the cases he mentioned in his short were not phony, although his quick explained them erroneously. Bacon reacted that there’s no “product difference” in between improperly explaining a genuine case and pointing out a phony one, as the guidelines about sincerity to the court use “with equivalent force” in any case.

“I didn’t understand that AI might hallucinate truths”

Aarons supplied a declaration when gotten in touch with by Ars today. He stated:

In March 2025 I consented to deal with an appeal and utilized ChatGPT to sum up the trial procedures. I composed the quick however the tabulation and the summary consisted of various mistakes. At the time, I didn’t understand that AI might hallucinate realities not just in my quick however in pleadings sent by other lawyers. I am delighted the court threw away my malfunctioning pleading and bought the general public protector to compose a brand-new quick on behalf of my previous customer. When it comes to myself, I am sorry however confident that the disciplinary board considers it was a sincere error. It is a lesson discovered for all experts who trust this effective however in some cases unsteady innovation.

Aarons informed the court that he started by utilizing Rev.com, a service that supplies AI records of audio files, and after that put that records and other products into ChatGPT.

“I think I did validate” that the trial records was precise, he stated when asked if he validated its precision before inputting it into ChatGPT. “The issue wasn’t in the records. It was when I packed it and all the other details” into ChatGPT, he stated. In addition to the records, Aarons stated he put “the record correct, the declaration of problems, and a few of the discovery” into ChatGPT.

Aarons stated he hoped the court would take actions to avoid other attorneys from making the very same error he did, such as by providing a standing order needing that briefs consist of a certificate of compliance concerning making use of AI. Justices dealt with Aaron’s recommendation as an interruption from the primary issue, his failure to validate the precision of the quick.

“Assume with me that you had actually depended on the work of a first-year attorney that was working for you which they had actually simply made things up … and you signed it. You ‘d remain in the very same specific soup you are right now,” Bacon stated. “So the concept that, the tip in your rundown that since the court didn’t inform you at the time that you did this, ‘beware,’ it in some way eases you of commitment, falls on outright deaf ears due to the fact that the guidelines of expert conduct currently inform you what your responsibility is.”

“Your customer is the one suffering”

“The other thing that’s entirely missing out on from your action is anything about what this has actually done to your customer,” Bacon went on to state. “That’s who I’m anxious about. You have actually provided rundown to the court that we can not count on, and your customer is the one suffering due to the fact that of this even more than you will ever suffer.”

Chief Justice Julie Vargas likewise informed Aarons, “I’m truly interested with the method you’re taking in this hearing. You appear to be informing all of us the policy that we’ve been considering for several years about what to do with AI, however you’re not speaking about how to resolve the scenario that’s in front people, which has considerable effect on a criminal accused who remains in custody, who’s going to remain in custody up until we solve this matter. And I truly do not care about the policy issues today. I appreciate what we’re going to finish with your customer and what we must carry out in this situation based upon the habits that you revealed us.”

Justice Michael Vigil stated there’s absolutely nothing incorrect with utilizing AI to assist compose a legal argument as long as the legal representative validates its precision. “It does not matter what the tool is. It does not matter how advanced the AI-generated program is or what enhancements they make, whatever,” he stated. “It does not matter whether you utilize a C-student legal representative or a A-student attorney [if] you didn’t inspect their work before you submitted the short. That’s the problem. And I want, or I’m hoping you would concentrate on that, take obligation for it.”

Aarons responded, “I hope I do not present myself as declining duty. I indicate, my only hope is that the errors that I made that in some way can avoid somebody else down the line from duplicating my error since the AI capability is incredible.”

Legal representative didn’t inform customer and household about ChatGPT utilize

Bacon slammed Aarons for not revealing the issue to his customer. “I have not. He’s a Spanish speaker and I wished to see what occurred today,” Aarons informed Bacon when she asked if he informed Sandoval about the issue.

Aarons stated he was employed by Sandoval’s bro and cousin, which he informed them there was an issue with a short. “They’re mindful that there was an issue with the very first short that I submitted which the court’s taking a look at that. That’s practically it,” he stated.

Bacon responded, “You didn’t state, ‘I count on ChatGPT and it sent lies to the court and the court seethes at me and now there’s a hearing.’ I’m thinking you didn’t inform your customer’s member of the family that.”

“I didn’t discuss ChatGPT,” Aarons stated. “I simply stated there was an issue with the quick that I submitted.” Bacon informed Aarons that his customer required to understand the level of the issue so he can make a notified choice “about whether you’re the best individual to represent him.”

Bacon asked Aarons whether he was dealing with the case on a public protector agreement or as personal counsel. Aarons addressed that he was employed as personal counsel however had not been paid.

“I’m not exactly sure that’s going to take place,” Bacon responded.

Jon is a Senior IT Reporter for Ars Technica. He covers the telecom market, Federal Communications Commission rulemakings, high speed customer affairs, lawsuit, and federal government guideline of the tech market.

151 Comments

  1. Listing image for first story in Most Read:

Learn more

As an Amazon Associate I earn from qualifying purchases.

You May Also Like

About the Author: tech