The Lawletter Vol. 51 No. 5
Amy G. Gore, Senior Attorney
As of October 2025, OpenAI's ChatGPT had approximately 800 million weekly active users. Jennifer Sor, Sam Altman Touts ChatGPT's 800 Million Weekly Users, Double All Its Main Competitors Combined, Business Insider (Oct. 8, 2025). Many of those users are attorneys and law firms.
With the use of this new technology comes the need to regulate how attorneys may and may not integrate that technology into a routine legal practice. Unsupervised use of artificial intelligence (“AI”) in legal matters implicates an attorney’s ethical duty of competence when not verified. Use of generalized AI and input of confidential client information run the risk of unauthorized disclosure to other users. See generally ABA Comm. on Ethics & Professional Responsibility, Formal Opinion 512. In yet another cautionary tale of the misuse of AI in the legal profession, the New Mexico Supreme Court held an attorney in direct contempt after he filed an appellate reply brief that had been generated by ChatGPT and contained hallucinated material. State v. Sandoval (In re Aarons), No. S-1-SC-40845, 2026 N.M. Unpub. LEXIS 16, at *1, 2026 LX 502549, 2026 WL 2661673 (Sept. 9, 2026). The material included not only misrepresentations of holdings from prior decisions, but also four wholly fabricated witnesses and the false testimony of several actual witnesses. The attorney admitted that he had used ChatGPT without verifying the results. To make his situation worse, the attorney failed to notify his client of the errors in the brief, or of the show cause proceedings brought on the court’s own motion.
The court found the attorney in direct contempt of court and referred the matter to the New Mexico Bar for further disciplinary consideration. The attorney was also barred from appearing before the court pending consideration by the Bar.
This type of irresponsible use of AI in the legal profession is becoming more and more common. Many of the cases involve hallucinated case citations. See, e.g., State ex rel. Okla. Bar Ass'n v. Reeves, 2026 OK 37 (unverified use of AI-generated false citations, contrary to firm rules, constituted “knowingly” making false statements to a tribunal and warranted public reprimand); State Nat'l Ins. Co. v. Treadwell, 826 F. Supp. 3d 1352, 1353 (N.D. Ala. 2026) (use of ChatGPT and submission to court of fabricated and unverified citations warrant public reprimand); Ruiz v. Magellan Fin. & Ins. Servs., No. CV-23-02090-PHX-DWL, 2026 U.S. Dist. LEXIS 190341, at *1 (D. Ariz. Aug. 24, 2026) (formal public reprimand was warranted by submission of three briefs with use of ChatGPT that generated false citations that were not verified before submission); In re Nwaubani, No. 25-9517, 2026 U.S. App. LEXIS 7501, at *1 n.1 (4th Cir. Mar. 11, 2026) (court sanctioned attorney with public admonishment for submitting brief with fabricated case citations). But at least one other case has involved a false citation to the record. See State v. Coleman, 2026-Ohio-965, ¶ 34, 280 N.E.3d 1042, 1059 (Ct. App.) (fabricated transcript quotations generated by an AI tool were submitted to the court).
Partners and senior attorneys who fail to supervise subordinate attorneys’ use of AI can face their own ethical sanctions. See Hill v. Workday, Inc., No. 23-cv-06558-PHK, 2026 U.S. Dist. LEXIS 93600, at *1 (N.D. Cal. Apr. 28, 2026) (supervising attorney’s failure to read brief containing AI-generated false citations warranted imposition of monetary penalty, mandatory CLE requirements, and circulation of sanctions order to all firm staff).
Not mentioned by the New Mexico court, but likely to be addressed with the New Mexico State Bar, is another one of the inherent dangers of the use of generative AI: breach of client confidences. Since generative AI systems continuously build upon the information that is inputted by attorneys, several jurisdictions have cautioned that confidential information may be inadvertently disclosed to others, and offered some guidance to protect client confidences. See New York City Bar Association Committee on Professional & Judicial Ethics – Decs (Aug. 7, 2024) (generative AI systems that keep inputted information entirely within the firm's own protected databases, sometimes called "closed" systems, do not present risk of disclosure of confidential information).
The generative function of ChatGPT could possibly account for the false testimony and fabricated witnesses created in the reply brief before the Sandoval court, making further investigation critical. It remains essential for all practitioners using these tools to scrupulously protect their client’s confidences and thoroughly review all results.
