Two federal district courts recently addressed the uncharted waters of the relationship between generative AI available on large language models and the attorney-client and work product privileges.
The United States District Court for the Southern District of New York recently ruled that AI outputs were not protected by the attorney client privilege, nor the work protect doctrine. See United States v. Heppner, No. 25 CR. 503 (JSR), 2026 WL 436479, at *2 (S.D.N.Y. Feb. 17, 2026). The Heppner case is a criminal matter where both parties were represented by counsel.
The Heppner court noted that the attorney client privilege protects “[c]ommunications (1) between a client and his or her attorney (2) that are intended to be, and in fact were, kept confidential (3) for the purpose of obtaining or providing legal advice.” Heppner, 2026 WL 436479, at *2 (internal citations omitted). In Heppner, the party’s use of AI failed all three elements of protected communications. First, the inputting party was a non-lawyer. Second, the communications were not confidential because AI is a “third-party,” and there are not “substantial privacy interests” in conversations with AI platforms. Id. Third, most (if not all) AI platforms expressly say their outputs should not be used as legal advice. Id. at *3.
Likewise, the court pointed out that the work-product doctrine protects the “mental processes of the attorney, providing a privileged area within which he can analyze and prepare his [or her] client’s case.” Id. (citing United States v. Nobles, 422 U.S. 225, 238 (1975)). The Heppner court held the AI outputs were “prepared by defendant on his own volition. That means that [defendant] was not acting as his counsel’s agent when he communicated with [the AI.]” Id. at *4.
The Court concluded its pioneering decision with a reflection of the discovery issues to come:
“Generative artificial intelligence presents a new frontier in the ongoing dialogue between technology and the law. Time will tell whether, as in the case of other technological advances, generative artificial intelligence will fulfill its promise to revolutionize the way we process information. But AI’s novelty does not mean that its use is not subject to longstanding legal principles, such as those governing the attorney-client privilege and the work product doctrine.“
Id.
Only one week earlier, the United States District Court for the Eastern District of Michigan denied-in-part a motion to compel AI generated documents. See Warner v. Gilbarco, Inc., 2026 WL 373043, at *3 (E.D. Mich. Feb. 10, 2026). In Warner, the Plaintiff was pro-se and the Defendants were represented by counsel.
In Warner, defendants sought to compel production of plaintiff’s “documents and information concerning her use of third-party AI tools in connection with this lawsuit. . .” Id. The court denied this request, citing Federal Rule of Civil Procedure 26(b)(3)(A) that work product requires the existence of a document “prepared in anticipation of litigation or for trial by another party or its representative.” Id. (emphasis in original). The court noted that the information was not relevant, and “even if marginally relevant, [the request wa]s not proportional.” Id.
The court further allowed the pro-se plaintiff to assert the work product protection over the AI generated documents. Id. Although the defendants argued that plaintiff had “waived the work-protect protection by using ChatGPT,” the court noted that “the work-product waiver has to be a waiver to an adversary or in a way likely to get in an adversary’s hand.” Id. (collecting cases). See e.g., United States v. Am. Tel. & Tel. Co., 642 F.2d 1285, 1299 (D.C. Cir. 1980) (“while the mere showing of a voluntary disclosure to a third person will generally suffice to show waiver of the attorney-client privilege, it should not suffice in itself for waiver of the work product privilege.”). The court held that generative AI programs are “tools, not persons,” and therefore the work product privilege could not have been waived to a tool. Id. Not surprisingly, the court denied a request for a privilege log.
Both Heppner and Warner are consistent with the principle that generative AI products are not a person, but a tool. However, generative AI products cannot act as an attorney and cannot cloak the actions of the user with privilege, unless that user is a lawyer and acting as counsel. Although the use of AI seems to go against the “spirit” of the work product doctrine – which is to protect lawyers/human thoughts – it appears that if a lawyer or one acting as counsel is using the tool to develop their thoughts, it could be protected.
This is a double edge sword. There is an overarching issue in the United States with access to justice – people need cheaper ways to make the law work for them. People deserve representation, especially in criminal and family law matters. People deserve easier ways to understand the law. AI can help with some of these issues. And, AI is frequently unreliable in such a nuanced field, like the law, where much of what we do lives in “gray” areas subject to the exercise of human judgment about individuals’ motives and intent.
What can I do as a legal practitioner in these new and uncharted waters?
As the Heppner Court stated, “time will tell” the contours of this new and developing area of law. Nevertheless, based on these decisions, here are a few practice tips:
- Lawyers should be advising their clients that anything they put into AI and its outputs may be discoverable in litigation. A practical and easy way to do this is to include a warning in your existing retainer agreements. Explicitly advise clients to refrain from putting counsel’s privileged advice through a public AI tool.
- If a client is insistent on using AI, have the client send you their question, and put the prompt into the generative AI product yourself. This way, the work-product doctrine has a chance of protecting the information.
- Lawyers should be considering AI prompts and outputs as a potential source of discoverable information from opposing parties and include these kinds of requests in their requests for production.
Lydia Rytting
moc.nqr@gnittyrl
Lydia Rytting is an associate in the firm’s Intellectual Property and Litigation Sections, where she represents clients in trademark, copyright, and trade secret disputes, including opposition and cancellation proceedings before the Trademark Trial & Appeal Board. She also prosecutes trademark applications and advises clients on brand management, protection, enforcement, and infringement matters, in addition to handling a range of commercial litigation matters such as contract disputes and business torts. She remains active in the Utah State Bar, Women Lawyers of Utah, and the LGBT & Allied Lawyers of Utah, and holds a bachelor’s degree in art history from the University of Utah.