Heppner and Gilbarco: Courts Apply Privilege and Work Product Protection to Generative AI Tools

Key Takeaways

• The U.S. District Court for the Southern District of New York recently held, in United States v. Bradley Heppner, that communications by a criminal defendant with a publicly available generative AI platform are not protected by attorney-client privilege or work product privilege.
• The court found no attorney-client relationship between the individual and the AI platform, no reasonable expectation of confidentiality, and no protected legal advice.
• Client-generated materials created without direction from counsel do not qualify as work product, whether generated by hand or generated by AI tools.
• In contrast, the U.S. District Court for the Eastern District of Michigan held, in Warner v. Gilbarco, Inc., that a pro se plaintiff's use of generative AI tools in connection with litigation did not waive work product protection, because AI platforms are “tools, not persons” and disclosure to them is not disclosure to an adversary.
• The decisions show courts beginning to apply the law of attorney-client privilege and work product doctrine to generative AI and, thus far, viewing the tasks and their outcomes as neither expanding nor contracting the protections long recognized under existing frameworks.

United States v. Bradley Heppner (S.D.N.Y.)

On February 17, 2026, the Southern District of New York, in United States v. Bradley Heppner, held that a criminal defendant's written exchanges with a “publicly available AI platform” are not protected by attorney-client privilege or work product doctrine and, thus, could be inspected by the government. As our previous Update explained in more detail, the court held that a criminal defendant’s self-initiated use of a publicly available generative AI platform to prepare documents outlining his potential defenses, which he later shared with counsel, were not privileged because (1) the platform was not a lawyer; and (2) the platform further trained on user inputs and stated in its privacy policy that it could share the information with third parties, including the government. The court further held that the documents were not work product because the defendant had prepared them on his own without instruction from counsel.

Warner v. Gilbarco, Inc. (E.D. Mich.)

Just one week before the Heppner ruling, a federal magistrate judge in the U.S. Court for the Eastern District of Michigan reached a different outcome on different facts—but applied the same principles to get there. In Warner v. Gilbarco, Inc., decided on February 10, 2026, the court denied the defendants’ motion to compel a pro se plaintiff to produce materials reflecting her use of generative AI tools in connection with the litigation. The decision illustrates the other side of the coin: Where the facts support the elements of the work product doctrine under well-settled law, generative AI tools do not disturb those protections.

Background

The plaintiff, proceeding pro se in an employment discrimination action, used a generative AI tool in preparing materials related to her case. The defendants sought to compel production of “all documents and information concerning her use of third-party AI tools in connection with this lawsuit.” They also asked the court to overrule the plaintiff’s attorney-client privilege and work product objections, arguing that her use of a generative AI tool waived any applicable protections. 

Work Product Doctrine Protects a Litigant's AI-Assisted Drafting

The court denied the defendants' motion on multiple grounds. As a threshold matter, the court found that the information sought was not discoverable, citing Rule 26(b)(3)(A)’s protection for materials “prepared in anticipation of litigation or for trial by or for another party or its representative” (emphasis added). Note that there is no requirement that a lawyer be involved in the creation of work product material. The court further found the request either irrelevant or disproportional under Rule 26(b)(1).

AI Tools Are 'Tools, Not Persons'

The court then squarely addressed the defendants’ waiver argument, drawing a critical distinction between the attorney-client privilege and work product protection. The court acknowledged the well-established principle that “voluntary disclosure of confidential material to a third party waives any applicable attorney-client privilege.” But it held that work product waiver requires something more: disclosure “to an adversary or in a way likely to get in an adversary's hand.” The court observed that “generative AI programs[] are tools, not persons, even if they may have administrators somewhere in the background.” Therefore, inputting litigation materials into a generative AI platform did not constitute disclosure to an adversary sufficient to waive work product protection.

Compelled Production Would Expose Protected Mental Impressions

The court agreed with the plaintiff that the defendants’ request improperly sought “internal analysis and mental impressions—i.e., her thought process—rather than any existing document or evidence,” and that compelling production of such materials “would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed.” Notably, the defendants’ waiver theory was “supported by no case law but only a Law360 article posing rhetorical questions,” and the court observed that “no cited case orders the production of what Defendants seek here: a litigant's internal mental impressions reformatted through software.”

The court characterized the defendants’ pursuit of the plaintiff's AI usage information as “a distraction from the merits” and described the motion as “a fishing expedition.”

Reconciling the Two Decisions

The Heppner and Gilbarco decisions, issued days apart, may at first glance appear to point in different directions—one stripping protection from AI-generated materials and the other upholding it. But the holdings are ultimately consistent and, taken together, illustrate that the arrival of generative AI has not changed the law governing privilege and work product protections. Both courts applied long-established principles to a new technology, and the differing outcomes flow from the differing facts.

In Heppner, the court found the attorney-client privilege inapplicable because the defendant communicated with a publicly available AI platform—not a lawyer—using a service whose privacy policy affirmatively disclaimed confidentiality and reserved the right to disclose user data to third parties, including government authorities. There was no attorney-client relationship, no expectation of confidentiality, and no direction from counsel. The work product doctrine likewise did not apply because the defendant acted entirely on his own initiative, and the materials did not reflect counsel's mental impressions or litigation strategy. These are straightforward applications of principles that long predate generative AI.

In Gilbarco, the court found that the work product doctrine protected a pro se litigant's AI-assisted drafting materials precisely because the doctrine shields a party's litigation-related thought processes regardless of the medium through which they are expressed. The court emphasized the well-established distinction between attorney-client privilege waiver—where voluntary disclosure to any third party may suffice—and work product waiver, which requires disclosure to an adversary or conduct likely to result in an adversary obtaining the material. Because generative AI platforms are “tools, not persons," using them to assist in drafting is no more a waiver of work product protection than using a word processor, a search engine, or a legal research database. Waiver requires disclosure to an adversary, which is unlikely to arise from mere use of a generative AI tool.

What is the same is that neither court proposed to create a new doctrine or carve out AI-specific exceptions. Instead, both asked the same questions courts have asked for decades: Was there a qualifying relationship? Was confidentiality maintained or reasonably expected? Were materials prepared in anticipation of litigation, by or at the direction of counsel, and reflective of counsel's mental processes? Generative AI tools simply present a new factual context in which to apply these familiar tests—and practitioners should expect courts to continue doing so.

Practical Implications

These two decisions together offer guidance for practitioners navigating the intersection of generative AI and privilege:

• Attorney-client privilege requires a human attorney. Communications with a generative AI platform are not communications with counsel. AI tools cannot form an attorney-client relationship, and no privilege attaches to exchanges between a client and a publicly available AI platform.
• Platform data practices matter to the attorney-client privilege. Using an AI platform that retains user inputs, trains on user data, and reserves the right to disclose data to third parties—including government authorities—might be viewed by a court as vitiating any reasonable expectation of confidentiality. Clients and counsel should carefully review platform privacy policies before using any AI tool in connection with a legal matter.
• Work product protection depends on the facts, not the tool. Materials prepared in anticipation of litigation by a party or its representative may retain work product protection even if generative AI was used in the drafting process. The key inquiries remain whether the materials were prepared by or at the direction of counsel and whether they reflect counsel’s mental impressions or litigation strategy.
• Work product waiver is harder to establish than privilege waiver. Unlike attorney-client privilege, which may be waived by voluntary disclosure to third parties outside the circle of privilege, work product waiver requires disclosure to an adversary or in a manner likely to reach an adversary. Using a generative AI tool—a tool, not a person—does not, standing alone, constitute such disclosure.
• Nonprivileged AI-generated outputs do not become privileged simply because they are later sent to counsel. The Heppner court was clear that transferring otherwise unprotected materials to a lawyer does not retroactively cloak them in privilege.
• Direction from counsel matters. In both cases, courts examined whether counsel directed the client's use of the AI platform. Materials created at counsel's direction stand on stronger footing under the work product doctrine; materials created independently by a client, without counsel's involvement, may lack the requisite nexus.

The overarching lesson is that generative AI is neither a magic wand that confers privilege nor a trap that automatically destroys it. The same principles that have governed attorney-client privilege and the work product doctrine for decades continue to apply—to telephone calls, memorandums, emails, and now to generative AI. Clients and counsel who understand and apply these familiar principles will be well positioned to navigate this evolving landscape.