As with many things and the law, as AI technology evolves, the laws surrounding technology and its use, and the ramifications of its use, are constantly being honed as unique factual circumstances present themselves. In a recent decision of first impression, United States v. Heppner, 25 CR 503 (JSR, 2026 BL 52143 S.D.N.Y. February 17, 2026) stands as a fundamental illustration of this evolution. The issue in the case concerned the discoverability of communications and reports generated through Heppner’s communications with Anthropic’s platform “Claude.” Heppner was an executive of several corporate entities, including a publicly traded company, who had been indicted by a grand jury for securities fraud, wire fraud, conspiracy to commit securities fraud, and falsification of corporate records, among other charges. He was arrested, and in connection with his arrest, the FBI executed a search warrant at his home, obtaining numerous documents and electronic devices. Among the seized materials were thirty-one (31) documents that memorialized communications that Heppner had with Claude. Heppner had initiated and prepared these reports after the indictment was unsealed and before his arrest.
The issue before the Court was the claim of attorney-client privilege or work product asserted by Heppner’s counsel. The argument was that the reports outlined defense strategy and potential arguments with respect to the facts and allegations anticipated to be brought by the government. Heppner had included in his chats with Claude information he had learned from counsel and had prepared the reports to speak with counsel and obtain legal advice. One can imagine the types of admissions and questions Heppner likely asked Claude.
In dispensing with his counsel’s argument for privilege and work product, the Court succinctly noted that AI documents are not and cannot be communications between Heppner and his counsel. Privileged communications, the Court held, require, among other things, “a trusting human relationship.” The Court noted that no such relationship exists, or could exist, between an AI user and a platform such as Claude. The Court also noted that the documents were not confidential because they had been expressly communicated to a third party, i.e., Claude. The Court noted that in order to use Claude, Heppner had consented to their privacy policy, which expressly advised that Claude “provides that Anthropic collects data on both users’ inputs” and Claude’s “outputs” and that such data is used to “train” Claude, and that Anthropic reserves the right to disclose such data to a host of “third parties.” As such, AI users cannot have a substantial privacy interest in conversations with AI platforms that reserve the right to share and disseminate such information and retain it in the normal course of business. Accordingly, the Court had no trouble referring to Claude as a third party and finding that confidentiality could not apply to such communications. As to Heppner’s counsel’s argument that the communications were for the purpose of obtaining legal advice, the Court also swiftly dispensed with it, noting that, when asked whether it could provide legal advice, Claude responded: “I’m not a lawyer and can’t provide formal legal advice or recommendations.”
Summing up the decision, the Court ruled that none of the communications or reports generated by Claude in response to Heppner’s inquiries, or Claude’s outputs, could be protected under the attorney-client privilege. As to the work product doctrine claim asserted by Heppner’s counsel, which is typically applied to protect the mental processes of the attorney as he prepares a client’s case, the Court noted that because the Claude reports were not prepared or generated by attorneys themselves the protection could not extend to the reports, because Heppner had created them on his own in conjunction with Claude, and they therefore could not contain content or insight into defense counsel strategy.
A last-ditch argument raised by Heppner’s counsel was that the AI documents should be protected because they incorporated information he had conveyed to Mr. Heppner during his representation. The Court found, however, that even if that was the case, and that the information input into Claude was privileged, the privilege was waived by sharing that information with Claude, just as it would be as if it were shared with any other third party.
The takeaway from the Heppner decision will be far-reaching, not only in criminal cases but in civil cases as well. One can certainly expect that in any future litigation, discovery requests will inquire whether users use Claude, ChatGPT, Copilot, DeepSeek, or any other AI chatbots. It will also likely become routine discovery to request conversations and inquiries posed to these platforms and any reports or summaries generated therefrom. And before you think of simply deleting such data, it is also most probable that preservation letters sent in advance of, or in anticipation of litigation, will advise that holds be placed on any such information under threat of spoliation.
So, before sharing your most private thoughts, fears, and/or concerns with AI, the more prudent course is likely to contact counsel. Only then will your communications be privileged and protected, and your counsel can then advise and/or direct you as to what role AI should play, or not play, in your assessment of your situation.
Contact Mike Sandner at msandner@pselaw.com or 937.223.1130 to discuss AI use in your particular situation.
