Using AI for Legal Advice: Is it intelligent to use artificial intelligence for advice about the law?
By: James E. Meadows
We are in the midst of a great artificial intelligence (“AI”) revolution. Hardly a day goes by that we don’t hear in the media, from academia, or even from our own families about the coming impact of generative AI on our lives. Many prophesy great economic and social disruptions from the use of AI while other “true believers” see a second Renaissance filled with potential. The reality likely is somewhere in between those two extremes.
Whether our civilization’s mass adoption of AI is or is not a good idea is not something this article can answer. However, whether (and how) individuals should use generative AI such as Claude, Gronk, and ChatGPT for legal advice can be known. While AI does have some utility in legal work, great caution is required when using it and it cannot eliminate the need for a knowledgeable and experienced attorney. Multiple recent cases, both across the country and right here in Missouri, involve non-lawyers using AI for legal work with disastrous consequences.
Risk of Using AI for Legal Advice
There are two primary risks in using AI programs without a licensed attorney’s involvement. First, the attorney-client privilege and work-product doctrine, which traditionally shield sensitive information from disclosure, do not protect the confidentiality of information shared by non-lawyers most AI systems. Your use of an AI system can cause your confidential legal information to be open to opposing parties in litigation or to the government, including in criminal prosecutions. Second, many AI programs, such as Claude and ChatGPT, cannot be trusted to answer legal questions correctly. AI systems routinely “hallucinate” legal statutes and cases, often making up results and getting the law wrong. Even when AI refers to real cases or statutes, it often widely misunderstands what the court, legislature or agency is saying in the cited source about the law.
Waver of Attorney-Client Privilege and the Work Product Doctrine When Using AI
Courts have concluded that information provided to or from an AI system by a non-lawyer is not protected by the attorney-client privilege. The U.S. District Court for the Southern District of New York, one of the most watched courts in America, ruled on February 17, 2026 that the use of a generative AI (in this case, Claude) waives the attorney-client privilege and is not protected by the work-product doctrine in a criminal prosecution. See U.S. v. Heppner, 820 Fed. Supp., 3rd, 292 (S.D.N.Y. 2026). In Heppner, the criminal defendant used Claude AI to analyze various legal defenses and theories available to him. He started seeking legal help from Claude AI after he had learned a federal investigation was pending against him but before he had been charged with a crime. When he was eventually charged a federal felony, the FBI arrested him, raided his home and seized items from his home, including his computer.
Among the information the FBI discovered were 31 records that the defendant had generated using Claude AI. These records outlined his legal theories, possible defenses, and relevant facts of the criminal investigation against him, as well as the defendant’s own thinking about the allegations. Heppner had created the AI documents without informing his attorneys that he was doing so and without any direction from his attorneys. Heppner’s attorneys sought to get the information back and prevent its use during further investigations or at trial, claiming the AI documents were protected by both the attorney-client privilege and the work-product doctrine.
The judge rejected both arguments. With regard to the attorney-client privilege, the Court noted that the attorney-client privilege protects documents from disclosure only when 1) those communications are between a client and his or her attorney, are 2) intended to be confidential and were in fact kept confidential, and finally 3) for the purpose of obtaining or providing legal advice. Citing United States v. Mejia, 655 F.3d 126 (2d Cir. 2011). According to the judge, AI records fail to meet at least two of these requirements and likely fail the third (for the purpose of legal advice) as well.
An AI system is not an attorney and direct attorney-client communication is considered core to the protection afforded by the attorney-client privilege. Also, information created or provided to an AI system is generally not, by contract, confidential. Most commercial AI programs allow information submitted to it to be shared for research and training purposes. Both of these factors prevent any protection under the attorney-client privilege.
Similarly, the court held the work product doctrine does not protect the AI-created records from disclosure. The work-product doctrine protects materials prepared by or at the direction of an attorney in anticipation of litigation or for trial. Heppner, 820 Fed. Supp., 3rd at 298. Documents created using AI, even if prepared in anticipation of litigation, are not prepared at counsel’s direction and do not reflect counsel’s mental processes or strategy, a key reason the work product doctrine exists. Id at 299. Only one other federal court, after Heppner, has addressed the work-product doctrine in connection with AI use by a non-lawyer. That case, Morgan v. V2X, Inc., 25-CV-1991, 2026 WL 864223 (D. Colo. 2026), did theoretical provide some protection under the work product doctrine but, as a practical matter, greatly limited the non-lawyer’s ability to use any AI system for legal task.
In Morgan, the plaintiff, who was proceeding without an attorney, was requested to give his former employer information about what AI program he was using. The employer also sought a court order prohibiting information involved in the litigation from being shared with a commercial AI system.
While the court noted that the work product doctrine could, in some circumstances, protect information shared by a non-attorney litigant with an AI system from disclosure to the opposing party, it issued a protective order barring the sharing of any information with any AI system that segregate information submitted for legal purposes from being shared with any third parties. The Court recognized that this likely prevented a non-lawyer party from being able to use any AI system, where attorneys could continue to do so through specialized AI products like Westlaw’s Co-Counsel or Harvey, another system used by many larger law firms.
Common AI Systems Do Not Provide Correct Legal Information
Leaving aside the risk to confidentiality presented by using AI for legal advice, AI systems often get the law wrong. Numerous courts around the country have imposed fines on both people representing themselves and attorneys using AI systems to prepare legal arguments. Courts discover AI system were used because the cases, statutes, and rules cited by AIs do not exist, or, if they do exist, are completely misunderstood by the AI program.
Missouri is not immune to this trend. In Kruse v. Karlen, 692 S.W.3d 43 (Mo. App. 2024), the Missouri Court of Appeal dismissed an appeal filed by an individual representing herself and sanctioned her $10,000 in attorney’s fees that she had to pay to the opposing side due to the use of AI. The Court noted that, of the twenty-four cases cited by the AI program, only two were real. Both of those two cases were completely misunderstood by AI and stood for principles entirely different from those for which they were referenced. Troubling, the AI system made up cases names and citations that, on the surface, appeared genuine. Only when someone went to look for the cases was it discovered that cases didn’t exist.
The Court of Appeals noted that the “bogus citations” could not, “for any reason….be countenanced and represent a flagrant violation of the duty of candor” owed to the court. Id at 53. The submission of fake cases and authority “constitutes an abuse of the judicial system,” and, even though the opposing party didn’t ask for any sanctions, the Court used its discretion and inherent power to order that the non-attorney personally pay $10,000 of her opponent’s legal fees.
The Kruse case is not alone. In June 2026, the Missouri Court of Appeals again fined a non-attorney $10,000 for using AI to create legal arguments. Boatner v. Hickerson, 2026 WL 1542070 (Mo. App. 2026). In Boatner, the AI again made up several non-existent cases. Of the six cases cited by the non-attorney, four simply didn’t exist, while the other two were only mentioned by name but never discussed or used in the argument. The Court again sanctioned the non-attorney $10,000 of the attorney’s fees of the opposing party.
Courts may not always impose sanctions on a party that uses AI. The decision seems to be based on the amount of effort the opposing party spends dealing with the fictitious cases. In Jones v. Simploy, Inc., 698 S.W.3d 480 (Mo. App. 2024), the Court of Appeals again confronted a non-attorney who used AI to create his arguments. The Court refrained from imposing sanctions because the case was appeal from a workers’ compensation matter in which the state of Missouri, the defendant, had not hired an outside attorney. Because the state had not spent attorney’s fees responding to the fake citations, the Court of Appeals refrained from issuing a sanction but did dismiss the non-attorney’s case. Id at 486.
Conclusion
While AI is an incredible tool that can greatly help both attorneys and clients, it must be used with caution. Documents shared with most publicly available AI systems waive both the attorney-client privilege and the work-product doctrine, exposing information to the opposing party and the government. AI created documents can even be used against a non-attorney who creates them as part of a criminal prosecution. AI systems also pose a significant risk of “hallucinations” that get the law wrong and can lead to monetary fines from a court. So, before turning to AI to answer a legal question, contact your attorney.