Two Federal Courts Chart Diverging Paths on the Discoverability of LLM Interactions
Two federal court decisions issued within days of each other in February 2026 have set the stage for a significant and unresolved question in modern litigation: Does attorney-client privilege or the work product doctrine protect relevant exchanges with a generative AI platform from disclosure? In U.S. v. Heppner, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026), Judge Jed Rakoff ruled that a criminal defendant’s exchanges with an AI platform were protected by neither the attorney-client privilege nor the work product doctrine. A week earlier, in Warner v. Gilbarco, No. 2:22-cv-11854 (E.D. Mich. Feb. 10, 2026), the U.S. District Court for the Eastern District of Michigan reached the opposite conclusion, finding that a pro se plaintiff’s use of an AI tool in connection with litigation preparation was protected as work product. Taken together, these rulings underscore that the law governing privilege in the context of large language models is developing rapidly and with significant practical consequences for litigants and their counsel.
In Heppner, the defendant was indicted for securities fraud and related charges arising from allegations of misconduct stemming from his alleged role in defrauding investors of a publicly traded company out of more than $150 million. When FBI agents arrested Heppner in November 2025 and executed a search warrant at his home, they seized approximately 31 documents memorializing his communications with a generative AI platform. According to Heppner’s counsel, the communications took place in 2025, after Heppner had received a grand jury subpoena and after it was clear he was the target of a federal investigation. Without any direction from his attorneys, Heppner used the AI platform to prepare reports that “outlined defense strategy” and “what he might argue with respect to the facts and the law” in anticipation of a potential indictment. Heppner’s counsel subsequently asserted privilege over the documents, and the government moved for a ruling that they were not protected.
Heppner advanced three principal arguments in support of privilege: (1) he input into the AI platform information learned from his attorneys, (2) he created the documents for the purpose of speaking with counsel to obtain legal advice and (3) he subsequently shared the AI outputs with counsel. The government countered that the communications failed to satisfy the foundational elements of either the attorney-client privilege or the work product doctrine because, in short, an AI platform is not an attorney, the communications were not confidential and the materials were not prepared by or at the direction of counsel.
Judge Rakoff agreed with the government on all counts, identifying at least two, and likely three, independently dispositive deficiencies in the privilege claim. First, because an AI platform is not an attorney, the communications were not between Heppner and his counsel. Second, the communications were not confidential. The AI platform’s privacy policy to which users consent permits data collection, model training and disclosure to third parties, including “governmental regulatory authorities,” which is consistent with the approach regarding the discoverability of AI prompts in In re OpenAI, Inc., Copyright Infringement Litigation, 2025 WL 3468036 (S.D.N.Y. Dec. 2, 2025). Third, and in what the court called a “closer call,” Heppner did not communicate withteh AI platform for the purpose of obtaining legal advice. The AI platform itself disclaims providing legal advice, and counsel conceded it “did not direct [Heppner] to run [AI platform] searches.” Although Heppner’s counsel argued the exchanges were made for the “express purpose of talking to counsel,” the court held that the relevant inquiry is whether Heppner intended to obtain legal advice from the AI platform, not whether he later shared the outputs with his lawyers. Nonprivileged communications, the court stressed, are not “alchemically changed into privileged ones upon being shared with counsel.” However, the court did note that had counsel directed Heppner to use an AI platform, the platform “might arguably be said to have functioned in a manner akin to a highly-trained professional who may act as a lawyer’s agent within the protection of the attorney-client privilege” leaving open a potential pathway for privilege where AI use occurs under attorney supervision.
On work product, Judge Rakoff similarly ruled against Heppner. Even assuming the documents were prepared “in anticipation of litigation,” they were not “prepared by or at the behest of counsel,” and they did not reflect defense counsel’s strategy. Heppner’s counsel conceded that the documents “were prepared by the defendant on his own volition” and may have “affected” counsel’s strategy but did not “reflect” it at the time they were created. The court noted that the Second Circuit has “repeatedly stressed that the purpose of the doctrine is to protect lawyers’ mental processes.”
In comparison, in Warner, pro se plaintiff Sohyon Warner brought claims against her employer alleging discrimination based on race. The dispute that generated this order arose during discovery when the defendants moved to compel production of “all documents and information concerning [plaintiff’s] use of third-party AI tools in connection with this lawsuit.” The defendants also sought to overrule the plaintiff’s work product objections to those materials, arguing that any protection had been waived by inputting litigation materials into a public version of AI platform. It is worth noting that in October 2025, the court modified the existing protective order to provide that “any documents marked confidential shall not be uploaded onto any AI platform.” This earlier order arose from the defendants’ concern that the pro se plaintiff might upload confidential discovery materials to an AI tool, which could compromise confidentiality.
The Warner v. Gilbarco decision reached a different conclusion from Heppner regarding application of the work product doctrine. The court rejected the defendants’ argument that the plaintiff’s use of a third-party AI platform effectively waived any work product protection, finding the AI-generated materials were protected under the work product doctrine per Federal Rule of Civil Procedure 26(b)(3)(A), which shields “documents and tangible things that are prepared in anticipation of litigation or for trial by another party or its representative.” Critically, the rule protects materials prepared “by another party or its representative” and not solely materials prepared by or at the direction of an attorney. A pro se litigant is the party, and the materials she prepares in anticipation of litigation qualify on the plain text of the rule. The court found that what the defendants really sought was “Plaintiff’s internal analysis and mental impressions (i.e., her thought process) rather than any existing document or evidence.” The court rejected the discovery request as a “fishing expedition” based on “speculation about what might exist in Plaintiff’s internal drafting processes” rather than any demonstrated relevance or need. 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.” Notably, the court characterized the defendants’ theory as one that, “if accepted, would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed.”
The divergence between Heppner and Warner reflects the fact-specific nature of the inquiry but also a deeper tension in the law. Both cases make clear that the use of AI tools for research, analysis and strategic planning in connection with litigation and investigations is both becoming more common and carries real discovery and privilege risks. As such, there are specific questions to consider. First, whether use of generative AI in connection with pending or anticipated litigation or investigations should occur at the direction and under the supervision of counsel, given the court’s suggestion in Heppner that attorney-directed use may support a privilege claim. Second, are litigation holds, preservation protocols and protective orders updated to address use of AI-generated content, including prompt inputs, platform outputs and any locally saved records of AI interactions. Third, do employees and clients understand that communications with public AI platforms are not confidential and should not be treated as substitutes for privileged communications with attorneys. Fourth, have organizations reviewed the terms of service and privacy policies of the AI platforms they use, paying particular attention to data retention, training and third-party disclosure provisions, which were all factors the Heppner court found weighed heavily against any expectation of confidentiality. Finally, practitioners should monitor developments closely, as the differences between Heppner and Warner are indicative of the fact-intensive inquiry that courts will apply and the evolution of definitive guidance will continue as these tools become more widely used.
