Federal Court Rules that a Litigant’s AI Materials Are Protected Work Product

Last week, I wrote about Judge Rakoff’s ruling in United States v. Heppner, in which the Southern District of New York held that a criminal defendant’s AI-generated legal research was not protected by the attorney-client privilege or the work product doctrine. Judge Rakoff, who has since issued a written memorandum, described the case as presenting “a question of first impression nationwide.” It was not the last word on the subject.

On February 10, 2026, the same date as the Heppner ruling, Magistrate Judge Anthony P. Patti of the Eastern District of Michigan issued an order in Warner v. Gilbarco, Inc., No. 2:24-cv-12333, addressing the same core question: are a litigant’s AI-generated litigation materials discoverable? Magistrate Judge Patti reached the opposite conclusion, ruling that the materials are protected under the work product doctrine and that using a consumer AI tool does not waive that protection.

Factual Background

Warner is an employment discrimination case. Plaintiff Sohyon Warner, proceeding pro se, brought claims against Gilbarco, Inc. and Vontier Corporation. During discovery, Defendants moved to compel production of “all documents and information concerning [Plaintiff’s] use of third-party AI tools in connection with this lawsuit.” Defendants also sought to overrule Plaintiff’s work product objections to those materials, arguing that any privilege had been waived by inputting litigation materials into ChatGPT.

The Court’s Ruling

The court denied Defendants’ motion to compel, finding that Plaintiff’s AI materials were not discoverable. The analysis proceeded on three grounds.

First, the court held that the materials were protected by the work product doctrine under Federal Rule of Civil Procedure 26(b)(3)(A), which shields from discovery “documents and tangible things that are prepared in anticipation of litigation or for trial by another party or its representative.” The court cited its own prior decision in Anderson v. Furst, No. 2:17-12676, 2019 WL 2284731 (E.D. Mich. May 29, 2019), for the proposition that a pro se litigant may assert work product protection over litigation materials. This is consistent with the plain text of Rule 26(b)(3), which protects materials prepared “by another party or its representative,” not merely those prepared by or at the direction of an attorney.

Second, the court rejected Defendants’ argument that Plaintiff waived work product protection by using ChatGPT. The court drew on the well-established distinction between waiver of the attorney-client privilege and waiver of work product. Attorney-client privilege can be waived by voluntary disclosure to any third party. Work product waiver, however, requires a disclosure to an adversary or in a manner likely to reach an adversary. See In re Columbia/HCA Healthcare Corp. Billing Pracs. Litig., 293 F.3d 289, 306 n.28 (6th Cir. 2002); 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 then went further, holding that “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.” This framing sidesteps the entire terms-of-service problem that contributed to the Heppner result. If an AI platform is a tool, like a word processor or legal research database, then inputting information into it is not a “disclosure” to a third party in any meaningful sense, and the waiver framework is never triggered.

Third, the court found that even if the AI materials were somehow discoverable, they were neither relevant nor proportional under Rule 26(b)(1). The court agreed with Plaintiff that what 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 admonished that “Defendants’ preoccupation with Plaintiff’s use of AI needs to abate.”

The Split with Heppner

Warner and Heppner now represent a genuine disagreement between two federal district courts on two distinct questions.

The first is waiver. Judge Rakoff in Heppner emphasized that Anthropic’s privacy policy destroyed any reasonable expectation of confidentiality, reasoning that information inputted into Claude “is not confidential.” Judge Patti in Warner addressed the waiver question from an entirely different angle: because work product waiver requires disclosure to an adversary—and because AI tools are “tools, not persons”—the terms of service are beside the point.

The second, and more fundamental, question is whether work product protection requires attorney involvement at all. In his written memorandum, Judge Rakoff expressly rejected the holding of Shih v. Petal Card, Inc., 565 F. Supp. 3d 557 (S.D.N.Y. 2021), which had held that work product protection applies to materials a plaintiff prepared in anticipation of litigation regardless of whether her attorney directed the work. Rakoff reasoned that extending the doctrine to materials created without attorney involvement “undermines the policy animating the work product doctrine,” which, as the Second Circuit has repeatedly stressed, is “to protect lawyers’ mental processes.” See In re Grand Jury Subpoenas, 318 F.3d 379, 383 (2d Cir. 2003); Matter of Grand Jury Subpoenas, 959 F.2d 1158, 1166 (2d Cir. 1992) (the doctrine “generally does not shield from discovery documents that were not prepared by the attorneys themselves, or their agents”).

Judge Patti in Warner—and the courts in Shih and Nichol v. City of Springfield, No. 6:14-cv-1983-AA, 2015 U.S. Dist. LEXIS 169901 (D. Or. Dec. 18, 2015)—read the doctrine differently. The text of Rule 26(b)(3) protects materials prepared “by another party or its representative.” As the Nichol court recognized, the plain language encompasses four categories: materials prepared (1) by a party; (2) by a party’s representative; (3) for a party; or (4) for a party's representative. In Nichol, a non-attorney friend who helped the plaintiff prepare for litigation was held to have created protected work product; in Shih, the plaintiff herself did. The absence of attorney direction may heighten the evidentiary burden of proving that materials were prepared “in anticipation of litigation,” but under the plain text it is not a categorical bar. The disagreement between Heppner and Warner thus reflects a deeper tension: whether work product’s scope is governed by the rule’s text or by its animating purpose as construed by the Second Circuit. That question will likely require appellate resolution.

The Takeaway

Warner is significant on two fronts. The “AI is a tool, not a person” holding, if adopted by other courts, would mean that a litigant’s use of AI to develop litigation strategy does not waive work product protection, regardless of what the AI platform’s terms of service say about data use. And the court’s implicit endorsement of the plain text of Rule 26(b)(3)—protecting materials prepared “by a party,” not just by or at the direction of attorneys—stands in direct opposition to Judge Rakoff's reading of the doctrine, which relied on the underlying purpose of the Rule. The practical upshot is that the work product doctrine may provide a more durable shield for AI-generated litigation materials than the attorney-client privilege, which remains vulnerable to the no-confidentiality rationale of Heppner. But which reading of the doctrine prevails will determine whether that shield extends to represented clients who use AI on their own initiative (the precise fact pattern that Heppner presented).