Pittsburgh, PA
Firm Alert
(by David White, Marc Felezzola and Angela Harrod)
Given the sharp rise in AI usage, courts have begun wrestling with the extent to which usage of AI tools for assistance with legal issues is protected from disclosure in discovery or otherwise. Early court decisions demonstrate there is considerable risk that communications between a client and an AI platform may not be protected by the attorney-client privilege or work-product doctrine.
The attorney-client privilege generally protects communications between a lawyer and client from disclosure. Similarly, per Federal Rule of Civil Procedure 26(b)(3)(A), the work-product doctrine protects “documents and tangible things that are prepared in anticipation of litigation or for trial.”
Two courts recently released opinions that provide important insights into the risk of accidentally waiving attorney-client privilege or work-product doctrine when a client turns to AI tools. In United States v. Heppner, No. 1:25-cr-00503-JSR, ECF. 27 (S.D.N.Y. Feb. 17, 2026), the court ruled that an individual’s inputs and the resulting outputs generated by a non-enterprise AI tool (meaning a public tool that is generally available at a consumer-level) are not protected by the attorney-client privilege or the work-product doctrine even if the individual using the AI tool was involved in litigation and was seeking legal advice.
The Heppner court held that the attorney-client privilege does not extend to “communications” between an individual and an AI platform, only to communications between a client and its counsel. The court further noted that the AI tool used by the individual in Heppner included a disclaimer that user submissions were not confidential. As such, the court held that the use of the tool constituted a third-party disclosure, which is not protected by attorney-client privilege.
The Heppner court also held that the AI-generated materials were not protected as work product because they were (1) not created by or at the direction of counsel, and (2) were not generated to reflect a legal strategy, especially since counsel was not involved in the AI use. It is unclear if the court would have ruled differently if counsel had instructed its client to use the AI tool.
The court also determined that the use of the AI tool independent of counsel is not work product as the protection only applies to “attorney” work product. The court declined to extend the protection to the client’s own work product.
There have been other cases in which courts have extended the work-product doctrine to include what clients generate for themselves. For example, in the same jurisdiction as Heppner, the court in Felder v. Warner Bros. Discovery, No. 23-Cv-8487, 2025 WL 3628224 (S.D.N.Y. Dec. 15, 2025) specifically held that client-generated non-AI materials are protected as work product.
The court in Heppner did not address whether counsel’s use of an AI tool to inform case strategy would allow the application of attorney-client privilege and/or the work-product doctrine. However, another recent case addressed just that issue.
In Warner v. Gilbarco, No. 2:24-cv-12333, 2026 BL 43591 (E.D. Mich. Feb. 10, 2026), the court held that materials generated by AI tools in litigation preparation were not discoverable. Unlike in Heppner, where the court held that disclosure to a consumer AI tool constituted a disclosure to a “third-party,” thereby destroying the attorney-client privilege, the court in Warner held that generative AI is a tool and not a person, and therefore, disclosure to it did not destroy confidentiality.
The Warner court found no meaningful distinction between using AI tools to prepare case materials and other actions that are historically protected as work product, such as an attorney taking notes during interviews of fact witnesses. The court did not even require a privilege log to be issued specific to the AI-generated materials, ruling that the party could simply object on the basis of attorney-client privilege and/or work-product doctrine.
Because the Heppner court took special notice that the generative AI tools were used by a non-attorney and were not confidential, clients and their counsel must be mindful of what AI tools are being utilized and by whom. Under Heppner, use of consumer-level AI tools such as ChatGPT, even if used by an attorney, could destroy attorney-client privilege. Additionally, as Heppner suggests, there may be no work-product protection for the inputs or outputs from an AI tool used by a non-attorney, even if the user is turning to the tool for assistance in formulating litigation strategy. While Warner suggests some courts may be more protective of AI-generated information, the prudent approach to protecting AI inputs and outputs from disclosure requires using tools that are specifically designed to maintain confidentiality. Even then, the usage of AI tools should be by or, at the very least, at the direction of counsel.
For questions about using AI-generative tools in legal matters and their protection under the attorney-client privilege or work-product doctrine, please contact David White at (412) 394-5680 or dwhite@babstcalland.com; Marc Felezzola at (412) 773-8705 or mfelezzola@babstcalland.com; or Angela Harrod at (412) 394-5688 or aharrod@babstcalland.com.
