United States v. Heppner (AI chats and privilege)
No. 1:25-cr-00510 (S.D.N.Y. Feb. 17, 2026) (Rakoff, J.)
A defendant asks a chatbot for legal strategy and the government reads the transcript
- In one lineEn una línea
- A client's own conversations with a public AI tool are not privileged or work product when no lawyer sought or directed them.Las conversaciones propias de un cliente con una herramienta pública de AI no son privilegiadas ni work product cuando ningún abogado las pidió o dirigió.
- RuleRegla
- Attorney-client privilege protects confidential communications made to a lawyer (or the lawyer's agent) to obtain legal advice; a person's exchanges with a generative AI tool, made on his own initiative and not at counsel's direction, are neither communications with counsel nor documents prepared by or for counsel in anticipation of litigation, so neither protection applies. The court suggested the analysis could differ where lawyers direct the use of the tool as part of the representation, by analogy to Kovel agents.
- Why this case is herePor qué está aquí este caso
- Assigned for class 10 with ABA Opinion 512 and Texas Opinion 705: the case every AI policy now cites for the rule that a public chatbot is a third party, not a lawyer, and that privilege follows the lawyer's direction, not the subject matter.
- Procedural posturePostura procesal
- Pretrial ruling in a criminal case on whether the defendant's conversations with a public generative AI tool, and the documents he produced with it, were shielded from the government by attorney-client privilege or the work product doctrine.
Facts
The defendant in a federal criminal case used a public generative AI tool on his own to analyze his situation and draft materials about his case, then shared some of the output with his lawyers. The government obtained the materials and the defendant claimed privilege and work product protection.
Procedural posture
Motion practice before trial in the Southern District of New York; the court ruled the materials unprotected.
Issue
Whether communications between a represented defendant and a generative AI tool, made without counsel’s involvement, are protected by the attorney-client privilege or the work product doctrine.
Holding
No. The defendant was not communicating with a lawyer for the purpose of obtaining legal advice, and the AI tool was not acting as the lawyers’ agent; the documents were not prepared by or for counsel in anticipation of litigation in the sense the doctrine requires.
Reasoning
Privilege attaches to the relationship with counsel, not to the subject matter of a document; a chatbot is a third party whose terms of service make the exchanges neither confidential nor part of the representation. The court noted the result might differ if the lawyers had directed the client to use the tool as part of their work, much as accountants or translators retained by counsel can fall inside the privilege.
Rule (own words)
If you want AI work protected, counsel must own it: lawyers choose the tool, direct its use, and keep it inside the engagement. Anything a client types into a public chatbot on his own is discoverable.
Class notes
Pair with ABA Opinion 512 (competence, confidentiality, consent) and Texas Opinion 705; the AI policy drafted in the week 10 lab should state the Heppner rule in one sentence for employees.
Say the answer out loud before revealing each part. Rate yourself honestly; the case comes back sooner when you miss.Di la respuesta en voz alta antes de revelar cada parte. Califícate con honestidad; el caso vuelve antes si fallas.
Panic kit: “Could you repeat the question, please?” · “Let me make sure I have the facts right: …” · “I'm not certain, but I'd argue … because …”Kit de pánico: “Could you repeat the question, please?” · “Let me make sure I have the facts right: …” · “I'm not certain, but I'd argue … because …”