AI chat logs not privileged

A federal judge in New York ruled that chats a criminal defendant had with Anthropic’s Claude were not protected by attorney-client privilege or the work-product doctrine. (mindingyourbusinesslitigation.com) Judge Jed Rakoff of the Southern District of New York made the ruling from the bench on February 10, 2026 and issued a written memorandum on February 17, 2026 in *United States v. Heppner*, a securities and wire fraud case against former GWG Holdings executive Bradley Heppner. (mindingyourbusinesslitigation.com) Rakoff wrote that the case appeared to present a “question of first impression nationwide” on whether communications with a publicly available AI platform, made during a criminal investigation, are shielded from government review. He answered that question “no.” (mindingyourbusinesslitigation.com) Attorney-client privilege protects confidential communications between a client and a lawyer for legal advice. Rakoff’s ruling turned on two points: Claude is not a lawyer, and sending information to a third-party AI service can break confidentiality. (foxrothschild.com) The court also rejected work-product protection, which usually covers materials prepared by lawyers or at lawyers’ direction for litigation. Crowell & Moring said Rakoff found the prompts and outputs were created by Heppner using a public AI tool, not by counsel, even though he later shared them with his defense team. (crowell.com) The facts were narrow. Crowell & Moring said Heppner used Claude after receiving a grand jury subpoena, generated about 30 documents combining his prompts and Claude’s responses, and those files were later seized from an electronic device during a search of his home. (crowell.com) Law firms have treated the decision as a warning about consumer AI tools, not a blanket rule for every internal system. Fox Rothschild said the opinion focused on a “public AI tool,” and several firm alerts have told clients to assume that anything typed into public chatbots may be discoverable. (foxrothschild.com; crowell.com)) That distinction matters for companies that store prompts, outputs, and audit trails. Orrick said saved chats, exports, and notes that memorialize AI conversations may be seized or subpoenaed, and companies should not assume privilege or work-product protection applies just because a lawyer later reviews them. (orrick.com) The ruling does not say every AI-related communication is unprotected. Adams and Reese said the opinion addressed exchanges with a publicly available generative AI platform, and lawyers are now steering clients toward tighter controls around what gets entered into consumer chatbots and how long those logs are retained. (adamsandreese.com) Heppner pleaded not guilty on November 10, 2025, after an indictment unsealed on November 4, 2025, and Rakoff’s February memorandum said trial was set for April 6, 2026. The immediate lesson from the opinion is simpler than the doctrine: a chatbot is not your lawyer. (mindingyourbusinesslitigation.com)