Legal Updates

Recent Court Decisions Highlight Privilege Waiver Issues Relating To AI Use

With the advent of artificial intelligence (“AI”) over the past few years, courts are beginning to confront how the attorney-client privilege and the attorney work-product doctrine apply when litigants use public, third-party generative AI tools to research, analyze, or draft materials related to their cases.

Two recent federal court decisions—one from the Southern District of New York and another from the Eastern District of Michigan—reach different conclusions on whether AI-related materials are protected by the attorney-client privilege and/or the work-product doctrine, based on platform terms, the role of counsel, and the nature of the materials. These cases provide helpful guidance for employers and their counsel on potential discovery of AI-related materials, privilege risks, and practical steps for managing AI use in litigation and day-to-day operations.

The Heppner Case

In United States v. Heppner, in the Southern District of New York, a federal grand jury issued an indictment charging the defendant, Bradley Heppner, with securities and wire fraud. During his arrest, the government seized devices that contained 31 AI generated documents from Anthropic’s platform Claude, which Heppner said he created “for the purpose of obtaining legal advice.” Defense counsel conceded that Heppner created these materials at his own initiative, not at counsel’s direction.

The government moved for a ruling that the AI generated documents were not privileged or protected work product. On February 10, 2026, Judge Jed S. Rakoff granted the motion from the bench.

A week later, Judge Rakoff issued a written decision holding that the documents prepared by Heppner using generative AI were not protected by attorney client privilege or the work product doctrine. Judge Rakoff held that the attorney-client privilege was waived through the defendant’s sharing of information with a third party AI platform. He also rejected the argument that the AI-generated materials fell within the attorney work product doctrine, noting that the AI platform was not legal counsel and the materials were created independently by the defendant, rather than at counsel’s direction.

It is noteworthy that the court treated the AI platform Claude as an external third party, and not as an internal agent, such as an interpreter. Additionally, Judge Rakoff emphasized that the platform’s terms and conditions specify that user data may be produced to third parties, thereby undercutting any claim of a reasonable expectation of privacy.

Finally, the court rejected the argument that the defendant’s sharing of the AI-generated materials with his counsel allowed for a “retroactive” claim of attorney-client privilege. Judge Rakoff observed, however, that had counsel directed Heppner to use Claude, the tool might arguably have functioned like a “highly trained professional” acting as the lawyer’s agent, in which case the work-product doctrine might have applied.

A Different Take: The Gilbarco Case

A week before the Heppner decision, a U.S. magistrate judge in the Eastern District of Michigan reached a different outcome.

In Warner v. Gilbarco, a pro se plaintiff alleging employment discrimination used a “Chat” AI tool to draft filings and conduct legal analysis. During discovery, the employer sought information about the plaintiff's use of third-party AI tools in connection with the lawsuit, including detailed questioning at her deposition. The plaintiff objected, arguing, in part, that the materials were protected by the work-product doctrine.

The magistrate judge hearing the case denied the employer’s motion to compel on the basis that the information sought by the employer was not discoverable under the Federal Rules of Civil Procedure. Specifically, the magistrate judge held that the pro se plaintiff’s AI-related materials were protected work product because they were prepared in anticipation of litigation and reflected her mental impressions, and that the defendants’ discovery requests improperly sought her internal thought processes rather than discoverable evidence.

The magistrate judge rejected the argument that the plaintiff had waived work-product protection by using an AI tool. In the magistrate judge’s view, such a waiver requires disclosure to an adversary (or making materials accessible to an adversary), and an AI chat feature is appropriately viewed as a tool, not as a person. This approach differs from Heppner’s view of AI as essentially a person for privilege purposes, based on the platform’s terms and conditions.

Key Takeaways From Heppner And Gilbarco

Based on these two recent decisions, there are a number of important points for employers and their counsel to consider:
 

• Treat AI use as potentially discoverable. AI prompts and outputs may be discoverable, much like internet searches, and courts may order broad discovery of AI “conversations.” Parties should anticipate that litigation may involve interrogatories, document requests and deposition questions seeking AI prompts and outputs, with the possibility of court intervention should the parties disagree as to the appropriateness of such discovery.

• Use of public AI may defeat privilege claims. A litigant’s independent use of public facing generative AI to assess legal exposure may not be protected by attorney client privilege or work product, even if intended to facilitate discussions with counsel, or shared with counsel at a later date.

• Platform nature and counsel direction matter. By contrast, enterprise tier AI platforms generally include confidentiality protections, thereby strengthening potential privilege arguments. Similarly, and as the Heppner decision signaled, counsel directed use of AI may be more likely to be treated as a privileged agent relationship.

• Evaluate internal practices and tools. Managers and HR employees should be cautioned not to use public AI platforms in making employment-related decisions or drafting correspondence. In addition, employers should assess, with input from counsel, their use of any automatic meeting transcription/summary features (e.g., Zoom AI Companion), as these types of AI-generated materials may be discoverable in litigation. They also can be incomplete or inaccurate or may memorialize off-the-cuff remarks that can later be misconstrued during litigation.

* * *

If you have questions about these recent court decisions or other legal issues relating to AI use, please reach out to one of our litigation attorneys.