Can a Client’s AI Searches be Discovered in Litigation?
When a potential client arrives with what appears to be an AI-generated description of their case—often including a detailed timeline, witness list, and an explanation of why they believe they should prevail—attorneys may reasonably wonder whether those AI searches could later become subject to discovery. Recent federal court decisions illustrate that the answer is not yet settled. Two courts addressing related circumstances reached different conclusions, suggesting that practitioners should approach the issue with care and provide thoughtful guidance to clients about their use of AI tools.
On February 10, 2026, federal judges in two separate matters considered whether a party’s use of generative AI tools could be examined in discovery. In Warner v. Gilbarco, Inc. (E.D. Mich.) Case No. 2:24-cv-12333, the court held that the plaintiff’s AI searches were protected from disclosure. By contrast, in United States v. Heppner (S.D.N.Y.) Case No. 25-cr-00503, the court determined that a criminal defendant’s AI communications were not protected by the attorney-client privilege or the work product doctrine.
Warner v. Gilbarco, Inc. (E.D. Mich.) Case No. 2:24-cv-12333
Warner is an employment discrimination action in which the plaintiff, at the time of the relevant discovery dispute, was proceeding without counsel. Magistrate Judge Anthony Patti of the Eastern District of Michigan denied a motion to compel seeking “[a]ll documents and information concerning her use of third-party AI tools in connection with this lawsuit.”
The court concluded that the requested material was protected under the work product doctrine. First, the court recognized that a self-represented litigant may invoke work product protections. The doctrine is not limited to attorneys and may apply when a party prepares materials in anticipation of litigation.
Second, the court explained that work product protection is generally waived only when the material is disclosed to an adversary or disclosed in a way that substantially increases the likelihood an adversary will obtain it. This standard differs from the attorney-client privilege, which can be waived through voluntary disclosure to many types of third parties.
The court also noted that generative AI programs are tools rather than persons. As Judge Patti observed, “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.” Because the plaintiff’s AI interactions reflected her internal analysis and thought processes in preparing for litigation, the court concluded they constituted protected work product.
United States v. Heppner (S.D.N.Y.) Case No. 25-cr-00503
The court in Heppner reached a different result. A grand jury had returned an indictment charging Bradley Heppner with multiple fraud-related offenses. After his arrest, the FBI executed a search warrant and seized electronic files, including documents containing exchanges between Mr. Heppner and “Claude,” a generative AI tool developed by Anthropic.
Mr. Heppner conducted these searches after it had become clear that he was under investigation. The searches were performed on his own initiative rather than at the direction of counsel. According to the record, the exchanges produced reports outlining possible defense strategies and potential arguments related to the facts and applicable law.
Mr. Heppner’s attorneys asserted both attorney-client privilege and work product protection. The government moved for a ruling that the documents were not protected, and Judge Jed Rakoff of the Southern District of New York agreed.
The court first held that the attorney-client privilege did not apply because the communications were between Mr. Heppner and an AI system, not between a client and a lawyer. The court also determined that Mr. Heppner could not reasonably expect the communications to remain confidential. Anthropic’s privacy policy indicated that user inputs and outputs could be used to train the system and could potentially be disclosed to third parties in connection with claims, disputes, or litigation.
Judge Rakoff further concluded that the communications were not made for the purpose of obtaining legal advice from counsel, since they were not undertaken at the suggestion or direction of an attorney. The court noted that circumstances might differ if a client’s communication with a non-lawyer occurred at the direction of counsel in order to facilitate legal advice. See, for example, United States v. Kovel, 296 F.2d 918 (2d Cir. 1961).
The court also declined to extend work product protection. Because the AI interactions were not prepared by or at the direction of counsel and did not reflect an attorney’s litigation strategy, the court concluded that the doctrine did not apply—even though the defendant’s attorneys later reviewed or considered the material.
Reconciling Warner and Heppner
Neither decision is binding on courts in California, but the reasoning in each case highlights factors that may influence future rulings.
A key distinction is the role of counsel. In Heppner, the defendant independently consulted an AI system without guidance from his attorneys. In Warner, the plaintiff was representing herself, meaning there was no separation between the party and counsel. The court therefore treated the AI interactions as part of the plaintiff’s own litigation preparation.
This distinction may be particularly relevant when prospective clients conduct AI research before retaining an attorney. Under reasoning similar to that in Warner, such searches could potentially be viewed as part of a party’s preparation for litigation. However, when a represented client independently consults an AI system to explore legal theories or litigation strategy, courts may analyze the issue more closely along the lines discussed in Heppner.
Practical Considerations for Attorneys
As of this writing, California courts have not directly addressed these questions. In the meantime, attorneys may wish to consider practical steps to reduce uncertainty surrounding client use of AI tools.
Discuss AI use early. Attorneys may wish to inform clients and prospective clients that their independent searches involving legal strategy or case analysis could potentially become discoverable in litigation.
Provide clear instructions when AI-assisted research is appropriate. If counsel asks a client to gather information or conduct research, written instructions describing the purpose of the work and directing the client to maintain confidentiality may help clarify that the work is being performed at counsel’s request.
Evaluate firm policies on AI tools. Law firms may benefit from establishing internal guidance on which AI platforms may be used, reviewing each platform’s privacy practices, and adjusting available settings that affect data retention or model training where possible.
Ask about prior AI use. Attorneys may also consider asking clients whether they have already used AI tools to analyze the dispute or develop potential legal arguments. Reviewing those materials early may help counsel determine whether additional steps—such as seeking a protective order—are appropriate.
Courts are only beginning to address how generative AI may affect doctrines such as attorney-client privilege and work product protection. This article is intended for informational purposes only and does not constitute legal advice. Individuals with questions about these issues should consult an attorney to discuss their specific circumstances.
