United States v. Heppner

United States v. Heppner, decided February 17th, 2026. Judge Rakoff, in the Southern District of New York, held that documents a criminal defendant created by conversing with a publicly available generative-AI platform — Anthropic's Claude — about his own case were protected by neither the attorney-client privilege nor the work-product doctrine. The court described the question as one of first impression nationwide. Here's the brief.

Bradley Heppner was an executive of several companies, including the publicly traded GWG Holdings. In October 2025, a grand jury indicted him for securities fraud, wire fraud, conspiracy, making false statements to auditors, and falsifying corporate records — charges that he defrauded GWG's investors of more than one hundred fifty million dollars.

When the FBI arrested him and searched his home, it seized about thirty-one documents memorializing written exchanges Heppner had with Claude. By his counsel's account, Heppner created them in 2025 — after he received a grand jury subpoena and knew he was the target — outlining defense strategy and the facts and law he expected the government to charge. He did so on his own, without any suggestion from counsel to use the tool.

Heppner asserted privilege over these AI documents. The parties segregated them under a protocol, and the government moved for a ruling that they were protected by neither the privilege nor the work-product doctrine. The court granted that motion from the bench on February 10th and issued this memorandum a week later.

The question: when a person communicates with a publicly available AI platform during a criminal investigation, are those communications protected by attorney-client privilege or work product?

The privilege, the court noted, is construed narrowly, and it requires three elements: a communication between client and attorney, intended and kept confidential, for the purpose of obtaining or giving legal advice. The work-product doctrine is related but distinct — it shelters materials prepared by or at the behest of counsel in anticipation of litigation, protecting the lawyer's mental processes.

On privilege, the court found the AI documents lacked at least two of the three elements. First, they were not communications with an attorney.

The court rejected the argument that an AI tool is just software like a word processor, reasoning that recognized privileges require a trusting human relationship with a licensed professional who owes fiduciary duties.

Second, the communications were not confidential. Under Anthropic's privacy policy, the court observed, user inputs and the model's outputs are collected, used to train the model, and may be disclosed to third parties — including governmental regulatory authorities.

Third, Heppner did not consult Claude to obtain legal advice: he used it of his own volition, not at counsel's direction, and Claude itself disclaimed giving legal advice. And sharing the documents with counsel afterward could not, as the court put it, alchemically turn non-privileged material into privileged material.

On work product, the court held the documents unprotected because they were not prepared by or at the behest of counsel and did not reflect counsel's strategy.

Declining to follow a district-court decision that had extended work product to materials a party prepared without an attorney's direction, the court grounded the doctrine in protecting the lawyer's own mental processes.

Heppner appears to be the first decision to address whether communications with a public AI platform during a criminal investigation are privileged, and it answers no. The court applied settled doctrine: an AI platform is not an attorney; a provider's data-use terms can defeat any expectation of confidentiality; and materials a client generates on his own initiative are neither privileged communications nor counsel's work product.

United States v. Heppner, 820 F. Supp. 3d 292, decided February 17th, 2026. I'm John Goodhue. Thanks for watching.