- within Technology topic(s)
- with Inhouse Counsel
- with readers working within the Law Firm industries
Thorny issues regarding whether AI generated documents are protected by the attorney client privilege or work product doctrine were recently highlighted in two divergent federal court cases. On February 10th, 2026, Judge Jed S. Rakoff of the U.S. District Court for the Southern District of New York ruled in U.S. v. Heppner1 that documents generated by the defendant’s use of an AI-generative tool were neither subject to attorney client privilege nor confidential work product. The documents, seized by the FBI in connection with a search warrant of the defendant’s home, included 31 documents generated by Anthropic’s AI-generative tool “Claude.” The defendant generated the documents after receiving a grand jury subpoena, related to potential fraud charges against him, for the purpose of outlining a defense strategy which he later shared with his counsel.
On the same day, a court in the Eastern District of Michigan issued an order in Warner v. Gilbarco, Inc.2 addressing the question of AI as work product and arrived at the opposite conclusion. Magistrate Judge Anthony P. Patti ruled that materials generated by a similar generative-AI as Claude, in this case ChatGPT, were prepared in anticipation of litigation and thus protectable work product and that using an AI tool did not waive that protection.
At the heart of these cases is the question of to what degree counsel’s involvement with AI constitutes protected communications and/or work product, and how independently generated AI materials fit within those categories.
United States v. Heppner
In United States v. Heppner, the defendant argued that the documents seized were protected by claims of privilege since the defendant had:
- input information into Claude that he had learned from counsel;
- created the documents for the purpose of speaking with counsel to obtain legal advice; and,
- shared the contents of the documents with counsel.
Of note, the defendant’s counsel conceded that that they “did not direct [Heppner] to run Claude searches.”3
The court ruled that the documents were neither subject to attorney client privilege nor work product doctrine. The decision rested on an analysis of the three defining elements of attorney client privilege namely communications (1) between a client and his or her attorney (2) that are intended to be, and in fact were, kept confidential (3) for the purpose of obtaining or providing legal advice.”4 The court argued that the defendant failed to prove that the AI-produced documents met these elements.
I. AI Is Not A Lawyer
On the first element, the court found that the documents were not communications between the defendant and his attorney, noting that defendant did not maintain that Claude was his attorney.
The court rejected the argument that Claude was the repository of attorney client communications analogous to Microsoft Word online or Dropbox Paper. The court noted that such applications are not privileged in and of themselves and, more importantly, privilege requires “a trusting human relationship” “who owes fiduciary duties and is subject to discipline.”5
II. AI Is Not Confidential
The Heppner court found that the 31 documents were not intended to be confidential. The court cited the written privacy policy of Claude, to which all users must consent to in order to utilize the AI. The privacy policy states that both users’ inputs and Claude’s outputs are subject to data collection and Anthropic reserves the right to disclose such data to third parties, including “governmental regulatory authorities.”6 Ultimately, even without a subpoena compelling it to do so, Anthropic may “disclose personal data to third parties in connection with claims, disputes[,] or litigation.”7 For these reasons, the defendant could have had no “reasonable expectation of confidentiality in his communications” 8 with Claude.
III. AI Cannot Provide Legal Advice
The court held lastly, that the defendant did not, and could not, communicate with Claude for the purposes of obtaining legal advice. The AI system disclaims providing legal advice and in response to a request states “I’m not a lawyer and can’t provide formal legal advice or recommendations” and that a user “should consult with a qualified attorney who can properly assess your specific circumstances.”9
In having found that the documents were not privileged at the time they were created, the court further held that they did not become privileged when defendant shared them with counsel, as it is well-established that non-privileged communications do not automatically become privileged just because they are sent to a lawyer.
IV. AI Does Not Constitute Attorney Work Product
In oral argument, the defendant cited to Federal Rule of Criminal Procedure 16(b) (2) (A),10 which provides that the Government may request pretrial discovery, and insofar as the defendant is obligated to satisfy the pretrial discovery requests, a defendant’s obligation does not extend to “discovery or inspection of reports, memoranda, or other documents made by the defendant, or the defendant’s attorney or agent, during the case’s investigation or defense.”11 The court rejected the argument on the ground that the Rule 16(b) did not apply because the AI documents were not requested by the Government and the defendant did not produce them in pretrial discovery, they were seized pursuant to a search warrant, the validity of which the defendant’s counsel does not contest.
The defendant’s reliance on Shih v. Petal Card, Inc.12 was also dismissed by the court. In Shih, a Magistrate Judge granted work product protection to communications between a plaintiff and an individual, who was at that time her lawyer, and later became her husband and which were made in anticipation of trial. The Heppner determined that in contrast, no lawyer was involved—plaintiff acted independently when he created the documents and they did not represent work product of his counsel.
Warner v. Gilbarco, Inc.
In Warner v. Gilbarco, Inc., a civil employment discrimination case, the defendants sought to compel production of “all documents and information concerning [Plaintiff’s] use of third-party AI tools in connection with this lawsuit.”13
Defendants sought to overrule Plaintiff’s work product objections to those materials, arguing that any privilege had been waived by the act of inputting litigation materials into ChatGPT.
The court denied the motion to compel, on the grounds that the information sought was non-discoverable work product, citing Federal Rules of Civil Procedure Rule 26(b)(3)(A): “Ordinarily, a party may not discover documents and tangible things that are prepared in anticipation of litigation or for trial by or for another party or its representative…”14
The court then addressed the waiver argument holding that plaintiff had not waived the privilege by submitting the materials to ChatGPT. According to the court, waiver only occurs when the material is made available to “an adversary or in a way likely to get in an adversary’s hand.”15
Judge Patti defined generative AI programs as “tools, not persons,” and reasoned that documents produced by such tools may form part of the plaintiff’s trial preparation even if counsel was not involved in the initial creation. In addressing work product, the Warner court reaffirmed that a “burden rests on the one who would invade that privacy to establish adequate reasons to justify production though a subpoena or court order,”16 a burden which the judge found was not by the defendants.
Take-Aways
At the center of these diverging opinions are two key issues: 1) the level of counsel’s involvement in, or direction of, their client’s actions; 2) categorization of AI as a “person” or a “tool.”
The Heppner court stated that if counsel had actually directed the generation of documents through AI, “Claude might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer’s agent within the protection of attorney-client privilege.”17
Warner v. Gilbarco, Inc. took a different tact. The court did not focus on whether counsel directed the plaintiff’s use of ChatGPT, but only the fact that the document was actually prepared in anticipation of litigation and became part of the trial strategy.
There will be undoubtedly more litigation on this issue. Until clearer guidance emerges, lawyers and litigants should approach the use of generative AI cautiously and with knowledge that if clients generate independent documents without the input or direction of counsel, they may have to be disclosed.
Footnotes
1. United States v. Heppner, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026).
2. Warner v. Gilbarco, Inc., No. 24CV12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026).
3. Heppner, 2026 WL 436479, at *2.
4. United States v. Mejia, 655 F.3d 126, 132 (2d Cir. 2011).
5. Heppner, 2026 WL 436479, at *2.
6. Privacy Policy, Anthropic (Feb. 19, 2025), https://www.anthropic.com/legal/archive/a2eecf43-807a-4a53-89dd-04c44c351138.
7. Id.
8. See Mejia, 655 F.3d at 132–34.
9. Heppner, 2026 WL 436479, at *3.
10. Fed. R. Crim. P. 16(b)(2)(A).
11. Id.
12. Shih v. Petal Card, Inc., 565 F. Supp. 3d 557 (S.D.N.Y. 2021).
13. Warner v. Gilbarco, Inc., No. 24CV12333, 2026 WL 373043, at *4 (E.D. Mich. Feb. 10, 2026).
14. Fed. R. Civ. P. 26(b)(3)(A).
15. See In re Columbia/HCA Healthcare Corp. Billing Practices Litig., 293 F.3d 289, 294–302 (6th Cir. 2002).
16. Hickman v. Taylor, 329 U.S. 495, 507–08 (1947).
17. Heppner, 2026 WL 436479, at *3.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]
