AI Searches and Documents May Not Be Protected by the Attorney-Client Privilege
A recent federal court decision from Judge Jed S. Rakoff of the US District Court for the Southern District of New York highlights a developing risk for lawyers and clients using generative AI tools in litigation. Materials created through the use of open AI platforms that reserve the right to disclose data to third parties may not be protected by the attorney-client privilege or the work product doctrine. This recent decision highlights the growing tension between traditional privilege doctrines and modern AI tools, particularly when third-party platforms are involved.
The Court’s Analysis: Be Cautious with Disclosure to a Third-Party AI Platform
In U.S. v. Heppner, the defendant was facing criminal charges arising from his alleged misconduct as an executive of several corporate entities. In connection with his arrest, the FBI seized numerous documents and materials pursuant to search warrants, including thirty-one documents that memorialized communications the defendant had with the generative AI platform “Claude.” Apparently, without any suggestion or direction from his counsel that he do so, the defendant prepared reports using Claude that outlined his potential defense strategy and defenses. When defendant’s counsel learned the documents were among those the government collected, he asserted privilege and demanded that they be segregated and not inspected pending resolution of the privilege claims.
After full briefing and a hearing, Judge Rakoff rejected the claim that the AI-generated documents were protected from disclosure, reasoning that because Claude explicitly tells its users that the data it collects, including both customer inputs and Claude’s outputs may be disclosed to third parties, the communications at issue were not confidential exchanges directly between attorney and client, but rather information voluntarily disclosed to a third-party. And because confidentiality is a foundational element of attorney-client privilege, the court found that disclosure to an external, non-confidential system undermined the claim.
In addition, although acknowledging it “presents a closer call,” the Court also found it important that the defendant did not consult with Claude “at the suggestion or direction of counsel.” Had counsel directed the defendant to use Claude, the AI platform 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 the attorney-client privilege.” Likewise, because the defendant communicated with Claude “of his own volition,” the communications do not merit protection under the work product doctrine because they did not reflect defense counsel’s strategy.
A More Nuanced Reality: How Privilege Was Lost
Importantly, the ruling is more nuanced than a blanket statement that “AI use destroys privilege.” According to the court’s analysis, it was critical that the defendant entered that information into an open, non-confidential AI platform. In addition, the documents were not generated at the request of counsel. The court concluded that neither attorney-client privilege nor the work product doctrine applied.
This decision leaves open two important questions: First, whether the outcome would have been different if the AI platform being used by the client was a confidential, or closed platform, rather than an open, non-confidential platform. And second, whether the privilege would have been lost if the lawyer had personally downloaded the information into the AI tool, or if the client had done so at the suggestion or direction of the attorney. Even then, however, the Court’s emphasis on the non-confidential nature of the platform suggests that privilege concerns remain.
This ruling also serves as a cautionary reminder that the attorney-client privilege depends on confidentiality and careful handling of communications. It also leaves open the question of the extent to which a court might extend privilege when some of the traditional notions of privilege are missing. For example, will a court protect an AI-generated transcript of a litigation strategy meeting? What about a recorded online meeting with a transcript and summary?
As lawyers and clients increasingly experiment with generative AI tools, they must evaluate whether those platforms preserve confidentiality or create the risk of waiver. Judge Rakoff’s concluding, cautionary remarks pinpoint the balance attorneys must strike when experimenting with AI tools and when advising clients on using AI: “AI’s novelty does not mean that its use is not subject to longstanding legal principles…”[1].
Conclusion
Organizations should proceed with caution and consider where information submitted to, or obtained from, AI tools could ultimately land. Confidential information should be kept that way, and AI may not provide the necessary protections and privileges. Our Litigation Practice Group is working on the cutting edge of these issues every day, and our attorneys can help with data privacy, AI usage and management policies, pre-litigation advising, training for team members, and so much more. If you or your organization have any questions about these or any other AI topics, please reach out to Pat Rooney, Shannon McDonough, or Hannah Schacherl Jansen.
[1] Memorandum, U.S. v. Heppner, 25-cr-00503-JSR, Feb. 17, 2026, at 12.
Related Attorneys
