Clients and lawyers alike are increasingly using generative artificial intelligence tools to organize facts, explore legal theories, and draft narrative summaries. But a recent court ruling calls into question whether conversations with AI should be protected in litigation, or whether they are fair game in discovery. This article will briefly explore the typical scope of discovery, documents that are considered privileged, and how the usage of AI fits into these frameworks. Lastly, we will review the Southern District of New York’s ruling in United States v. Heppner.
Scope of Discovery
Under both Virginia and federal law, the scope of discovery is very broad, generally covering all “nonprivileged” material that is “relevant to any party’s claims or defenses.” The discovery process often requires litigants to answer questions, and produce documents, reflecting or concerning all correspondence that is relevant to the claims, allegations or defenses in the lawsuit. This often includes correspondence between parties and non-parties, including emails, text messages, letters, etc.
Attorney-Client Privilege
Perhaps the most notable exception to the broad scope of discovery are the protections provided by the attorney-client privilege. Both Virginia and federal law protect confidential communications between a lawyer and their client if the client is seeking legal advice, which often covers most communications with counsel. Generally speaking, these communications are off limits in discovery.
There are exceptions to the attorney-client privilege (e.g., crime/fraud exception) and the privilege may also be waived by certain actions. Relevant to this article, the attorney-client privilege may be waived if the communications involve or are made in the presence of a third-party. For example, if a party and its counsel are exchanging emails regarding legal advice, but the email chain has copied in a third-party, the correspondence may be discoverable.
Work Product Doctrine
Under Virginia and federal law, the work product doctrine provides that documents and work product that are prepared for “in anticipation of litigation” or “for trial” are non-discoverable. The work product doctrine protects materials prepared by counsel in anticipation of litigation, designed to shield an attorney’s mental impressions, strategies, and legal theories. However, and similar to the attorney-client privilege, the work product doctrine can be waived if counsel shares its work product with a third-party (typically excluding experts hired by the party).
United States v. Heppner, No. 1:25-cr-00503-JSR (S.D.N.Y. 2026)
The Southern District of New York was recently tasked with analyzing how the aforementioned privileges coincide with the usage of AI, and if conversations with AI should be deemed discoverable.
In Heppner, the Defendant used, as many parties now do, the AI platform known as Claude (created by Anthropic) to outline his defense, discuss legal strategies, and seek legal advice. Part of Defendant’s usage of AI concerning his potential legal exposure pre-dated his arrest. The Government sought production of the communications with AI, but the Defendant argued that they were protected by the attorney-client privilege and the work product doctrine. The Defendant argued the communications were privileged because they involved information conveyed to the Defendant by his counsel and the conversations were conducted in anticipation of, or preparation for, litigation.
The Court rejected the Defendant’s arguments, finding that the communications will be subject to discovery. The Court concluded that any attorney-client privileges were waived because AI was considered a “third party” and because the conversations were inherently not “confidential.” Moreover, the conversations were not covered by the work product doctrine because AI is not legal counsel and the conversations did not involve legal counsel.
The Use of AI in Litigation Moving Forward
Litigants, potential litigants, and their attorneys need to be acutely aware that clients’ conversations with AI may be discoverable. A litigant using AI to test its legal theories or case weaknesses may now be akin to a criminal defendant who uses Google to ask, “how to dispose of a dead body?” If the communications, which are often far more candid and detailed than a Google search, may be discoverable by the opposing party, the consequences will be far reaching.
The ruling in Heppner will not be the end of these discussions, as the developments in case law are just beginning. AI platforms may respond by altering their platforms (e.g., making communications “confidential” in the eyes of the law). Or, perhaps, AI chats can be set up to include both attorneys and their client, to potentially maintain work product protections. We will continue to monitor these developments as courts grapple with the intersection of technology and privilege.
If you have questions about the usage of artificial intelligence in litigation or legal proceedings, please contact Harrison Clinton at (703) 526-5587 or hclinton@beankinney.com or Andrew Gregg at (703) 284-7254 or agregg@beankinney.com.
This article is for informational purposes only and does not contain or convey legal advice. Consult a lawyer. Any views or opinions expressed herein are those of the authors and are not necessarily the views of any client.