The twice-monthly “Dear Ethics Lawyer” column is part of a training regimen of the Legal Ethics Project, authored by Mark Hinderks, former managing partner and counsel to an AmLaw 200 firm. Read More

Q: Dear Ethics Lawyer,
I thought AI was supposed to make the provision of legal service more efficient and effective, to the benefit of both lawyers and clients. But now I hear about the Heppner decision from New York, and how the use of GenAI may destroy attorney-client privilege and work product. What do I need to know about this? What should I tell clients?

A: Judge Rakoff’s memorandum decision in United States v. Heppner, 25 CR. 503 (S.D.N.Y. Feb. 17, 2026) has certainly created a stir, but in the still-unsettled world about how interaction with GenAI tools affects legal matters it represents at most a relatively mainstream application of legal principles. In Heppner, the criminal defendant—unprompted by and independent of any attorney—did his own written exchanges with GenAI platform Claude operated by Anthropic after he learned that he was the target of an investigation. These exchanges and results were memorialized in 31 documents subsequently discovered and seized pursuant to an FBI search warrant. The defendant then asserted that these documents were protected by attorney-client privilege and the work product doctrine.

The court held that attorney-client privilege did not apply because the communications were not between a client and lawyer. It noted that Claude is not a lawyer and therefore could not fill that role for privilege purposes. It also rejected an argument that likened use of Claude to use of other cloud-based software tools, noting that no software tool creates a basis for privilege, there must be an attorney-client relationship with a licensed professional for any privilege to arise. It also found that the confidentiality requisite of privilege was missing, given disclosures in Claude’s privacy policy about the use of information for training purposes and potential disclosure to third parties for other purposes. The court also noted that Heppner communicated with Claude of his own volition, without direction of counsel, thereby undercutting the possibility that Claude could be considered an agent or tool of the lawyer. The fact that he later shared the results with his lawyer could not fix this original flaw; non-privileged communication cannot be made privileged retroactively by sharing them with an attorney.

The court similarly rejected work product protection for the documents because, even if they were prepared “in anticipation of litigation,” they were not prepared by or at the direction or request of counsel, and do not reflect or contain the strategy of mental impressions of counsel. As with the claim of attorney-client privilege, the lack of involvement of a lawyer at the time the documents were created proved fatal. The court rejected the reasoning of Shih v. Petal Card, Inc., 565 F. Supp. 3d 557 (S.D.N.Y. 2021), which held that communications with a lawyer in anticipation of litigation were protected regardless of whether the lawyer was acting as her counsel at the time, noting that the purpose of the work product doctrine is to protect lawyers’ mental processes, and that in Heppner’s case, he acted on his own and there was no lawyer involved.

While Heppner will not be the last word about privilege protection for AI-related communications and information, there are some key takeaways.

  • Open, non-enterprise GenAI systems that share or use information do not satisfy the confidentiality element of privilege, and both clients and lawyers should assume that communications with them may not be privileged.
  • Communications that are not made to or from a lawyer, not made to convey a lawyer’s advice or not or made to gather information to provide to a lawyer to support a request for legal advice are most likely not going to be privileged. Nor will analysis performed by or for a person who is not a lawyer and not acting at the direction of a lawyer be protected by the work product doctrine.
  • Advise clients that their own use of open, non-enterprise GenAI will most likely not be protected from disclosure, and that you will only use enterprise or “closed loop” GenAI. This protects the confidentiality of client information and your own analysis or strategy supporting the application of privilege and work product protection.
  • Put these disclosures in your engagement letter terms for all new matters and consider a broad message or bulletin concerning the use of GenAI to existing clients if you have not already done so. The Ethics Lawyer