Court: AI chats not privileged

A federal judge in New York ruled that a defendant’s chats with Anthropic’s Claude were not protected by attorney-client privilege and could be inspected by prosecutors. (harvardlawreview.org) The case is United States v. Bradley Heppner in the Southern District of New York, where Judge Jed S. Rakoff issued a written opinion on February 17, 2026, after a February 10 bench ruling. Heppner was indicted in October 2025 on fraud-related charges tied to GWG Holdings and pleaded not guilty in November 2025. (perkinscoie.com) (courtlistener.com) Federal Bureau of Investigation agents seized about 31 documents memorializing Heppner’s exchanges with Claude after his arrest. His lawyers said he used the chatbot after receiving a grand jury subpoena to prepare reports on defense strategy and then shared those reports with counsel. (harvardlawreview.org) (adamsandreese.com) Attorney-client privilege usually protects confidential communications between a client and a lawyer for legal advice. Rakoff said Claude was not a lawyer, the exchanges involved a third-party platform, and Anthropic’s policy said it could collect inputs and outputs, use them for training, and disclose them to third parties, including government authorities. (perkinscoie.com) (harvardlawreview.org) Rakoff also rejected work-product protection, which can shield material prepared for litigation. The court said Heppner created the Claude materials on his own, without direction from counsel, so they did not qualify. (adamsandreese.com) (perkinscoie.com) The ruling has pushed law firms to warn clients that chatbot conversations can be discoverable in criminal and civil cases. Reuters reported on April 15, 2026, that more than a dozen major United States firms had issued client alerts or contract language telling clients not to feed legal advice or sensitive facts into public AI systems. (usnews.com) Some lawyers are drawing a narrower lesson than the headlines suggest. Venable said commentary claiming any use of generative artificial intelligence automatically waives privilege “overstates” Heppner, because the opinion focused on a publicly available tool, a user acting without counsel’s direction, and a platform that did not promise confidentiality. (venable.com) Other firms are telling clients that the safest assumption is that anything typed into a consumer chatbot may later have to be produced. Fox Rothschild said the decision appears to be the first written federal ruling squarely addressing privilege claims for communications with a publicly available generative artificial intelligence platform. (foxrothschild.com) Heppner’s case is still pending, but the immediate effect is already visible in legal practice: public chatbot logs are being treated less like a lawyer’s office and more like material a court can demand. (usnews.com) (harvardlawreview.org)