Categories
A condensed version of this article originally appeared in Arkansas Business, authored by WLJ Labor & Employment attorney Troy Price.
The rapid development and widespread availability of artificial intelligence (AI) tools raises new legal questions and challenges for individuals and businesses that use these large language models. So far in 2026, several courts have considered the question of whether a party to a lawsuit may obtain access to information that the opposing party shared or generated while using AI. The question especially arises when a party to a lawsuit or its agents has currently or in the past used one of the broadly available AI platforms whose terms of service allow the vendor to make use of, or disclose, anything the user inputs, uploads, or generates. The issue is whether the use of a generative AI tool under these terms means the user has given up the right to keep an opponent in court from gathering information that otherwise could have been confidential.
The legal community quickly took note of United States v. Heppner, a February 2026 decision by a New York federal judge in a criminal securities fraud case. Legal observers knew that, even when criminal charges are not involved, the judge’s analysis in the case provided important lessons and considerations for business and individuals.
The dispute in the Heppner case arose after the FBI seized about thirty documents memorializing communications the defendant had with the generative AI platform “Claude.” Without his lawyers’ involvement, the defendant input into Claude information received from his lawyers, and he created AI documents for the purpose of speaking with counsel. The lawyers later tried to keep prosecutors from viewing or using the seized AI documents, asserting both the attorney-client privilege and the doctrine that shields information regarding an attorney’s mental processes when representing a client (the work product doctrine).
The court held that the defendant’s AI research on defense strategies was not protected by the attorney-client privilege because, among other things, Claude is not an attorney. The judge wrote that the attorney-client privilege has always required a “trusting human relationship” with a “licensed professional who owes fiduciary duties and is subject to discipline.”
Just as significant to the judge was the fact that the written privacy policy to which users of Claude agreed stated that both the user’s “inputs” and Claude’s “outputs” would be used to “train” Claude, and were also subject to being shared with “governmental regulatory authorities.” Considering the agreement of Claude users to this policy, the judge determined that the defendant had no reason to believe communications with Claude were confidential. The attorney-client privilege applies to communications with an attorney that are intended to be confidential.
Finally, the Claude AI platform specifically states that it does not give legal advice. The defendant in the case did not utilize Claude at his attorney’s request. Therefore, the judge reasoned, the situation was different from when a lawyer’s staff does research at the lawyer’s request. The mere fact that the defendant shared AI documents that were not legal advice with his lawyers did not convert those documents into legal advice protected by the attorney-client privilege.
Furthermore, the defendant’s lawyers were unable to keep prosecutors from reviewing or using his AI documents on “work product” grounds, the court decided, because the AI documents were not prepared “by or at the behest of an attorney” in anticipation of litigation, and therefore did not disclose defense counsel’s mental processes.
Because the FBI simply seized the AI documents in the Heppner case, there was no issue there about the civil litigation rules that require one party to turn over documents to the opposing party upon request. In civil cases the risk of waiving the attorney-client privilege remains a serious concern when any attorney-represented party inputs confidential communication into an AI platform. However, in Warner v. Gilbarco, Inc., a Michigan federal trial court upheld protection under the civil rules for a self-represented party’s AI-generated trial preparation materials. The court held that the plaintiff had not waived the protection of the rule by using AI, in part because “ChatGPT (and other generative AI programs) are tools, not persons.”
Similarly, in Morgan v. V2X, Inc, a federal trial court in Colorado pointed out that the work-product doctrine discussed in the Heppner criminal case protected only the mental processes of attorneys, while the federal civil rules provide protection for the trial preparation materials of parties and their representatives. (This is true for civil suits under the state court rules in Arkansas as well.)
In summary, the issues arising from use of generative AI by parties and potential parties in civil lawsuits are far from decided. What is clear is that inputting confidential communication from or to one’s attorney (including strategy plans and drafts of court papers) into an AI platform that may share or train on such information poses a grave risk to the attorney-client privilege. Even when no litigation is brewing at the moment, clients should be aware of the risk that using an AI platform on confidential communication with their lawyers may waive the attorney-client privilege for that exchange if a lawsuit is filed in the future. While every client should review the privacy policies of any AI platform in use, the best practice may be to avoid AI platforms completely when it comes to confidential communications.