As the use of artificial intelligence (AI) becomes more prevalent in day-to-day life and in the legal field, in particular, thorny questions arise regarding the implications of that use. One such question is whether exchanges with a publicly available generative AI platform in connection with pending litigation are protected by the attorney-client privilege or the work product doctrine. In a matter of first impression nationwide, U.S. District Judge Jed S. Rakoff of the Southern District of New York answered that question in the negative and required a defendant to provide the prosecution documents memorializing litigation-related communications with a generative AI platform.[1] Applying traditional principles governing the attorney-client privilege and the work product doctrine, the court reasoned that the communications did not involve an attorney-client relationship, were not confidential, were not made for the purpose of obtaining legal advice, and did not reflect an attorney’s trial strategy.[2] The ruling will likely impact whether legal protections are afforded to AI communications, prompts, and output in both litigation and regulatory inquiries, including state attorneys general (AG) investigations.
Background
On October 28, 2025, a grand jury indicted Bradley Heppner on a number of securities and wire fraud charges arising out of misconduct in his role as an executive of several corporations.[3] In the months leading up to the indictment, Heppner became aware that he was the target of the government’s investigation and, anticipating an indictment, began preparing his defense, including by retaining counsel. Heppner also utilized the generative AI platform Claude to prepare reports outlining his defense strategy and his potential arguments moving forward.[4] Importantly, Heppner did so of his own volition and “[w]ithout any suggestion from counsel.”[5] Heppner then provided those documents to his counsel.
On November 4, 2025, the FBI executed a search warrant at Heppner’s home in connection with his arrest.[6] Among the materials seized were 31 documents that memorialized the communications between Heppner and Claude. Heppner contended that those documents were privileged because the communications involved information Heppner learned from counsel and he had created them for the purpose of speaking with his counsel, to whom he ultimately provided the documents.[7]
The government thereafter moved to admit the documents as part of its case-in-chief asserting that the documents were not protected by either the attorney-client privilege or the work product doctrine.[8] The court heard oral argument on that motion on February 10, 2026, and orally granted the motion.[9] On February 17, 2026, the court issued a written memorandum explaining its order.[10]
Judge Rakoff’s Decision
In its written memorandum, the court held that Heppner’s documents failed to satisfy the well-accepted elements of the attorney-client privilege and the work product doctrine.
Attorney-Client Privilege
As to attorney-client privilege, the court concluded that the communications with Claude failed to satisfy any of the privilege’s three elements.[11]
First, the court determined that the communications did not involve an “attorney-client relationship” because neither Heppner nor Claude was an attorney.[12] And the absence of a “trusting human relationship” with a “licensed professional who owes fiduciary duties and is subject to discipline” alone negated the application of the attorney-client privilege.[13]
Second, the court concluded that the communications were “not confidential,” as is required to assert the attorney-client privilege.[14] Beyond the fact that Heppner communicated with a third-party AI platform, Claude’s user consent policy expressly states that it collects data on communications with Claude and that it reserves the right to disclose that data to third parties, including governmental authorities in connection with litigation.[15] Heppner therefore could not have had a “reasonable expectation of confidentiality in his communications.”[16]
Third, although a “closer call,” the court held that Heppner “did not communicate with Claude for the purpose of obtaining legal advice.”[17] While Heppner allegedly communicated with Claude for the “express purpose of talking to counsel,” he did not do so at the “suggestion or direction of counsel.”[18] The court posited that had counsel directed Heppner to utilize Claude, there may have been an argument that Claude acted as counsel’s agent within the protection of the attorney-client privilege. But because the communications were of Heppner’s “own volition,” the only question was whether Heppner intended to obtain legal advice from Claude. And, because he did not, the privilege did not apply.[19] Nonprivileged communications, the court reasoned, cannot be turned into privileged ones simply because they are shared with counsel.[20]
Work Product Doctrine
The court next held that the documents memorializing the communications between Heppner and Claude did not satisfy the elements necessary to invoke the work product doctrine, which protects “materials prepared by or at the behest of counsel in anticipation of litigation or for trial.”[21]
The court explained that, even if the documents were created in anticipation of litigation, they were “not prepared by or at the behest of counsel” and they did not reflect counsel’s trial strategy at the time they were created — though they affected counsel’s strategy moving forward.[22]
Key Takeaways
- The court’s decision is narrow. The court did not conclusively determine that the use of a third-party generative AI tool will always result in mandated disclosure of communications. Rather, the court’s decision is fact-specific, concluding only that the use of AI by a nonattorney that is not undertaken at the behest of an attorney is not covered by the attorney-client privilege or the work product doctrine. The question remains whether the work product doctrine can be invoked if an attorney — or a client at the behest of an attorney — utilizes AI to memorialize the attorney’s thought process in anticipation of litigation. Additionally, the court’s decision is only binding in the Southern District of New York’s federal courts. However, the order could serve as persuasive authority in other jurisdictions given it is a ruling of first impression.
- The ruling will have a regulatory impact. State AGs are increasingly active in the AI space, particularly given the Trump administration’s hands-off approach to AI governance. Indeed, AGs will often seek AI-generated data as part of civil investigative demands and other subpoenas. Given this, AGs are sure to cite the instant decision to pierce claims of privilege and work product as they seek AI communications and data through various processes. It is therefore incumbent upon counsel and businesses to use all due caution when interacting with AI platforms, and certainly if a matter may be subject to future regulatory inquiry.
- The appropriate use of AI in litigation is evolving. Throughout the memorandum, Judge Rakoff noted the novelty of the question presented and highlighted various questions that his opinion did not resolve, such as whether the outcome would have been different if Heppner’s counsel had directed Heppner to utilize Claude. Courts are likely to grapple with these questions in the immediate future. Counsel and businesses should thus keep abreast of AI-related rulings as variations of the instant factual scenario may result in significantly more impactful decisions.
- Be thoughtful when utilizing AI. As daily AI-use proliferates, its use will have consequences. And, in the context of legal proceedings, those consequences can be serious, such as the destruction of privilege surrounding AI communications. Publicly available generative AI platforms in particular pose great risk as information entered as prompts most often does not remain confidential and ownership of its output is often unclear. It is therefore important to thoughtfully consider whether the use of AI will help or hurt any litigation strategy or impact regulatory investigations, and experienced counsel should be involved in all decisions and actions involving AI use in legal matters.
[1] February 10, 2026 minute entry, United States v. Heppner, No. 25-cr-00506-JSR (S.D.N.Y.) (“Heppner“); Dkt. 27 at 1–2, Heppner.
[2] Dkt. 27 at 4–12, Heppner.
[3] Id. at 2.
[4] Id. at 3.
[5] Id.
[6] Id.
[7] Id. at 3–4.
[8] Id. at 4.
[9] February 10, 2026 minute entry, Heppner; Dkt. 27 at 4, Heppner.
[10] Dkt. 27, Heppner.
[11] Id. at 5–8.
[12] Id. at 5.
[13] Id. at 5–6.
[14] Id. at 6.
[15] Id.
[16] Id. at 7.
[17] Id.
[18] Id.
[19] Id. at 7–8.
[20] Id. at 8.
[21] Id. at 9.
[22] Id. at 9–10.
