Are conversations with artificial intelligence protected by the attorney- client privilege? A commentary on United States v. Heppner, 17/2/2026
By Albert Estrada and Clara Tarrés, members of Molins Investigaciones Internas
The widespread use of artificial intelligence (AI, hereinafter) in our daily lives and in the workplace is now a reality. The numbers speak for themselves: three years after its public release, one of the most well-known AI platforms has over eight hundred (800) million weekly users[1] .
The legal profession is no exception. In Spain, at least one lawyer[2] and one judge[3] have already been sanctioned for the improper use of these tools in the course of their work.
In the judgment rendered on February 10, 2026, by the United States District Court for the Southern District of New York in the case United States v. In United States v. Bradley Heppner, Judge Jed S. Rakoff expressly and exclusively resolved the following issue raised by the U.S. Government’s representatives in the context of the criminal proceedings against Mr. Bradley Heppner: whether the conversation that Mr. Heppner had with an AI tool to obtain legal advice about current or potential criminal proceedings against him is entitled to protection from government investigative activities under the attorney-client privilege or the work product doctrine.
Aware of the historic importance of his ruling- as it is the first to resolve this issue in the common law jurisdiction and, likely, in the world- Judge Rakoff thoroughly analyzes whether the specific case meets the conditions for the application of both doctrines.
The facts are based on three relevant circumstances: i) the privacy policy to which Bradley Heppner consented as a user of the AI platform states that data entered and received through the dialogue window may be used by the company that owns the platform to train the tool. The company also reserves the right to disclose the data to third parties, including public authorities; ii) Bradley used the tool on his own initiative , not on the advice of his attorney; iii) the conversation between Bradley Heppner and the AI was not guided by the advice or instructions of his attorney, but by his own considerations and reasoning.
Based on these facts, the judge concluded that Mr. Bradley Heppner’s conversations with the AI tool did not merit protection under either the attorney-client privilege or the work product doctrine.
First, regarding attorney-client privilege, the court found that the necessary requirements for granting protection to the conversations Bradley Heppner had with the AI were not met. On the one hand, the platform cannot be equated with an attorney or any other professional who is subject to duties of confidentiality, professional secrecy, and their own ethical standards. On the other hand, the platform’s own privacy policy, which the user accepts, allows the platform operator to access, use, and, in certain instances, share the information exchanged through the tool, which precludes any presumption of confidentiality.
With regard to the work-product doctrine, the Court similarly concluded that the documents generated from the AI’s interactions were not considered protected either. The exchanges were initiated solely by Mr. Heppner, and although they were related to an existing or potential criminal investigation, they did not reflect his defense strategy, much less any part of it. Consequently, the documents could not be considered to have been prepared by an attorney or by his client under the attorney’s direction. As such, they did not merit protection, and the conclusion was similarly unfavorable.
Although the judgment in question appears intended to establish legal precedent on the matter, the decision is, by definition, tied to the specific case. It cannot prevent another judge or court from reaching the opposite conclusion in significantly different circumstances. In fact, from a reading of the judgment itself, it is clear that if the privacy policy agreed upon by the parties had been different, and if the conversation had been initiated and, to some extent, guided by a lawyer, the conclusions could have been different. In fact, the first factor alone – the confidential conditions of the conversation between the user and the AI – could be sufficient, as notes taken by an individual for their own use, with the intention of delivering them to their attorney and requesting their assistance to defend them in a current or potential criminal case, are considered covered by the work product doctrine. (United States v. DeFonte, 441 F.3d 92, 95–96 (2d Cir. 2006 – per curiam)[4] .
We’ll see what the future holds for us. Given how rapidly AI is becoming increasingly prevalent in our lives, it will be sooner rather than later.