Divergent: Two Federal Courts Chart Seemingly Opposite Paths On Work Product Protection For AI-Generated Materials

Within the span of a single week in February 2026, two federal courts issued groundbreaking rulings on a question of increasing urgency for litigants and their counsel: are materials generated through interactions with publicly available artificial intelligence platforms protected from discovery under the attorney-client privilege or the work product doctrine? Comparing the two decisions, the answer appears to depend on who created the materials, why, and how those materials relate to the litigation strategy of the party or counsel. In United States v. Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 17, 2026), Judge Jed S. Rakoff held that a criminal defendant's exchanges with the AI platform Claude prior to counsel's involvement were not protected by privilege or the work product doctrine. Just days earlier, in Warner v. Gilbarco, Inc., No. 2:24-cv-12333 (E.D. Mich. Feb. 10, 2026), Magistrate Judge Anthony P. Patti reached the opposite conclusion in a case of a pro se plaintiff's use of publicly available ChatGPT in connection with her lawsuit, ruling that the prompts and outputs were protected from disclosure as material prepared in anticipation of litigation, or a party-litigant's work product. Both courts appear to have addressed issues of first impression, and neither identified any prior reported decision squarely on point.

United States v. Heppner: No Privilege, No Work Product Protection
Bradley Heppner, a former executive of a publicly traded company, was indicted in October 2025 on various charges related to securities fraud. FBI agents executed a search warrant at Heppner's home and seized numerous documents and electronic devices, including documents memorializing communications Heppner had with Claude, the generative AI platform operated by Anthropic. Heppner's counsel represented that these communications occurred in 2025, after Heppner had received a grand jury subpoena and after it was clear he was the target of the investigation. While the opinion does not expressly state when Heppner retained counsel, it appears that Heppner was represented by counsel at the time of engagement with Claude (for example, footnote 3 in the Heppner opinion mentioned that AI documents incorporated information that was conveyed by counsel to Heppner in the course of representation). Counsel further asserted that Heppner had, without any suggestion from counsel, "prepared reports that outlined defense strategy, that outlined what he might argue with respect to the facts and the law that we anticipated that the government might be charging." Heppner claimed privilege over these AI documents on three grounds: (1) he had inputted information learned from counsel into Claude; (2) he had created the documents for the purpose of speaking with counsel to obtain legal advice; (3) he subsequently shared the contents with counsel. The Government moved for a ruling that the AI documents were protected by neither the attorney-client privilege nor the work product doctrine.

Judge Rakoff's opinion methodically dismantled each element of Heppner's privilege claim. First, the AI documents were not communications between Heppner and his counsel. Claude is not an attorney, and, as the court observed, "in the absence of an attorney-client relationship, the discussion of legal issues between two non-attorneys is not protected by attorney-client privilege." The court rejected the argument that Claude should be treated like cloud-based word processing software, reasoning that such applications are not intrinsically privileged and that all recognized privileges require a trusting human relationship with a licensed professional who owes fiduciary duties and is subject to discipline.

Second, the communications were not confidential. Anthropic's written privacy policy, to which users of Claude consent, provides that Anthropic collects data on both users' inputs and Claude's outputs, uses such data to train Claude, and reserves the right to disclose such data to third parties, including "governmental regulatory authorities." The court concluded that Heppner could have had no reasonable expectation of confidentiality in his communications with Claude. The AI documents were therefore not like confidential notes that a client prepares with the intent of sharing them with an attorney, because Heppner first shared the equivalent of his notes with a third party.

Third, the court found that Heppner did not communicate with Claude for the purpose of obtaining legal advice. While Judge Rakoff noted that this presented a closer call, he emphasized that Heppner used Claude of his own volition, not at the suggestion or direction of counsel. The court noted that had counsel directed Heppner to use Claude, 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," citing the Second Circuit's Kovel¹ doctrine. Because Heppner acted independently, what mattered was whether he intended to obtain legal advice from Claude. The Claude platform disclaims providing legal advice, responding when asked that "I'm not a lawyer and can't provide formal legal advice." The court further held that non-privileged communications are not "alchemically changed into privileged ones upon being shared with counsel," and that because the AI documents would not be privileged if they remained in Heppner's hands, they did not acquire protection merely because they were transferred to counsel.

Turning to the work product doctrine, Judge Rakoff acknowledged that the doctrine "shelters the mental processes of the attorney, providing a privileged area within which he can analyze and prepare his client's case." However, the court found the AI documents undeserving of protection because, even assuming they were prepared in anticipation of litigation, they were not "prepared by or at the behest of counsel," nor did they reflect defense counsel's strategy.

Heppner's counsel admitted that the AI documents were prepared by the defendant on his own volition, meaning Heppner was not acting as his counsel's agent when he communicated with Claude. While counsel conceded that the AI documents did "affect" counsel's strategy going forward, they did not reflect counsel's strategy at the time Heppner created them.

Judge Rakoff also expressly declined to follow Shih v. Petal Card, Inc., 565 F. Supp. 3d 557 (S.D.N.Y. 2021), a decision from a Magistrate Judge in the same district that had concluded the work product doctrine is not limited to materials prepared by or at the direction of an attorney. Judge Rakoff reasoned that this conclusion undermines the policy animating the work product doctrine, which is to preserve a zone of privacy in which a lawyer can prepare and develop legal theories and strategy with an eye toward litigation. The court emphasized the Second Circuit's repeated holdings that the doctrine's purpose is not generally promoted by shielding from discovery materials in an attorney's possession that were prepared neither by the attorney nor his agents.

Warner v. Gilbarco: Work Product Protection Upheld
By contrast, Warner v. Gilbarco is a civil matter that arose from an employment dispute. Plaintiff Sohyon Warner was proceeding pro se. The case involved multiple rounds of contentious discovery disputes, including over defendants' demand for production of "all documents and information concerning [plaintiff's] use of third-party AI tools in connection with this lawsuit." Specifically, defendants argued that Warner's use of publicly available ChatGPT in connection with the litigation should be subject to discovery, and that any work product protection was waived by her use of a generative AI platform.

Magistrate Judge Patti denied the defendants' request, finding that the information sought was not discoverable under Federal Rules of Civil Procedure (FRCP) 26(b)(3)(A), which provides that "[o]rdinarily, a party may not discover documents and tangible things that are prepared in anticipation of litigation or for trial by or for another party or its representative[...]" The court further found that even if the information was marginally relevant, it was not proportional under FRCP 26(b)(1). On the question of whether work product protection applied, the court squarely held that a pro se litigant has the right to assert work product protection over such material, citing Anderson v. Furst, No. 2:17-12676, 2019 WL 2284731, at *4 (E.D. Mich. May 29, 2019).

As to whether the plaintiff waived work product protection by using publicly available ChatGPT, the court drew a critical distinction between attorney-client privilege and work product waiver. The court cited United States v. Am. Tel. & Tel. Co., 642 F.2d 1285, 1299 (D.C. Cir. 1980) for the proposition that "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 emphasized that work product waiver requires disclosure to an adversary or in a manner likely to result in disclosure to an adversary, citing In re Columbia/HCA Healthcare Corp. Billing Practices Litigation, 293 F.3d 289, 306 n.28 (6th Cir. 2002). Importantly, the court characterized "ChatGPT (and other generative AI programs)" as "tools, not persons, even if they may have administrators somewhere in the background."

The court agreed with the plaintiff's argument that the defendants' motion sought to "compel plaintiff's internal analysis and mental impressions — i.e., her thought process — rather than any existing document or evidence." Judge Patti also credited the plaintiff's warning that the defendants' theory, which was supported by no case law but only a Law360 article posing rhetorical questions, "would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed." The court admonished that "[d]efendants' preoccupation with plaintiff's use of AI needs to abate."

Can the Two Decisions Be Reconciled?
The Heppner and Warner rulings appear contradictory at first glance. The tension between them illuminates the doctrinal fault lines that future courts will need to navigate. If we look to reconcile the two decisions, the most important distinguishing factor may be the pro se posture of the plaintiff in Warner.

Because Ms. Warner represented herself, her use of AI tools to develop litigation strategy was, in a meaningful sense, the functional equivalent of an attorney using those same tools in case preparation. Her "internal analysis and mental impressions" were, by definition, the mental processes of the person directing the litigation, which is what the work product doctrine is designed to protect. In Heppner, by contrast, while the opinion does not expressly state when Heppner had retained counsel, the defendant was apparently represented by counsel but acted on his own initiative in connection with the use of AI, and the AI documents neither reflected counsel's strategy nor were created at counsel's behest. The result in Heppner likely would not have been different had he acted pro se in anticipation of litigation rather than being represented by counsel. Judge Rakoff's dictum suggests as much, as he expressly disagreed with Shih v. Petal Card and emphasized that the work product doctrine applies only to counsel's work—or, at most, to work performed at the direction of counsel.

The decisions also reflect different views on whether sharing information with an AI platform constitutes a meaningful "disclosure" for waiver purposes. Judge Rakoff treated Heppner's inputs to Claude as tantamount to sharing information with any other third party, emphasizing Anthropic's privacy policy and data practices. Judge Patti, by contrast, described generative AI programs as "tools, not persons," and applied the Sixth Circuit's more protective standard for work product waiver, which requires disclosure to or in a manner likely to reach an adversary. A different result in Heppner may have been possible had a paid version of an AI platform been used, particularly one governed by an agreement containing express no-training and non-disclosure provisions, including restrictions on human access by the AI platform vendor itself.

Lessons Learned for Lawyers and Litigants
These decisions offer several important takeaways for litigants, their counsel, and their opponents. First, proceed with caution when using AI platforms to analyze litigation strategy. The Heppner decision makes clear that sharing privileged or strategic information with a publicly available AI platform may destroy both attorney-client privilege and work product protection, at least in the Second Circuit. Communications with a publicly available AI tool are not confidential when the platform's terms of service permit data collection, model training, and disclosure to third parties.

Second, early involvement of counsel may change this analysis. Judge Rakoff expressly noted that had Heppner's counsel directed him to use Claude, the platform "might arguably" have functioned as a lawyer's agent under the Kovel doctrine. This dictum suggests that AI tools used under the direction and supervision of counsel may receive stronger privilege protections than tools used by a lawyer's client independently. On the other hand, pro se litigants may enjoy broader work product protections for AI-generated materials, at least in the Sixth Circuit. Because a pro se litigant is essentially both a client and her own attorney, her use of AI tools to develop case strategy may be inseparable from the mental processes the work product doctrine protects. For represented parties, the line between personal research and attorney-directed work product will be a critical distinction.

Third, litigants should review AI platform terms of service before use in any litigation-related context. Anthropic's privacy policy, which allows data collection and sharing with governmental authorities, was relevant to the Heppner court's finding that there was no reasonable expectation of confidentiality. For lawyers, given their duty of client confidentiality, making sure that their interaction with AI tools is strictly confidential is non-negotiable. Counsel should make certain to use enterprise or professional versions of AI platforms that offer the requisite privacy and security commitments, a caution that should be considered for any vendors used by counsel and/or their clients. The Heppner court's rationale could be reasonably applied beyond the AI context and acts as a reminder that counsel should vet vendors involved in their representation of clients.

Heppner and Warner decisions herald the first wave of judicial engagement with the privilege implications of generative AI in litigation. As Judge Rakoff observed, "[g]enerative artificial intelligence presents a new frontier in the ongoing dialogue between technology and the law," but "AI's novelty does not mean that its use is not subject to longstanding legal principles." The divergent outcomes in these two decisions underscore that the longstanding principles of the attorney-client privilege and the work product doctrine will be applied with close attention to the specific facts of each case. Courts, practitioners, and clients would be well-served to monitor this rapidly developing area of law as further decisions emerge.

Footnote

1 Articulated in United States v. Kovel, 296 F.2d 918 (2d Cir. 1961), the Kovel doctrine extends the attorney-client privilege to communications with certain non-attorney professionals, such as scientists, interpreters or accountants, when they are retained as agents of the attorney to assist in providing legal advice to the client.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.