Chatbot logs discoverable

A federal judge in New York ordered a fraud defendant to turn over documents he created with Anthropic’s Claude, rejecting claims that they were protected by attorney-client privilege. (mindingyourbusinesslitigation.com) Judge Jed S. Rakoff of the Southern District of New York first ruled from the bench on February 10, 2026, then issued a written opinion on February 17 in *United States v. Heppner*, No. 25 Cr. 503. The opinion said the answer to the privilege question was “no” when a user communicates with a publicly available artificial intelligence platform during a criminal investigation. (mindingyourbusinesslitigation.com) The case involved Bradley Heppner, who is charged with securities fraud, wire fraud, conspiracy, false statements to auditors, and falsifying records. After Federal Bureau of Investigation agents searched his home on November 4, 2025, they found about 31 documents that his lawyers said memorialized exchanges with Claude. (media.clarkhill.com) Attorney-client privilege usually protects confidential communications between a client and a lawyer made to get legal advice. Rakoff wrote that Claude was “not a lawyer” and that sharing information with Anthropic’s consumer chatbot was closer to disclosing it to a third party than confiding in counsel. (debevoisedatablog.com) Rakoff also rejected work-product protection, which can shield material prepared because of litigation. He wrote that Heppner created the Claude materials on his own, not at his lawyers’ direction, and compared the chatbot to an outside consultant who had no role in the defense team. (debevoise.com) The opinion has drawn attention because Rakoff described it as a “question of first impression nationwide,” meaning a federal court had not previously squarely answered it. Law firms have since circulated client alerts warning that prompts and outputs from public generative artificial intelligence tools may be discoverable in court. (mindingyourbusinesslitigation.com) Some lawyers have said the ruling is narrower than headlines suggest. Venable wrote that the decision turned on a public consumer tool, the lack of lawyer direction, and the record in this case, not on a blanket rule that every use of generative artificial intelligence destroys privilege. (venable.com) Other firms have read the opinion as a warning against putting legal strategy into public chatbots at all. Debevoise said the ruling “underscores the risks” of using consumer artificial intelligence platforms for legal research or case analysis without tighter controls. (debevoise.com) Anthropic was not accused of wrongdoing in the case, and the ruling did not ban lawyers or clients from using artificial intelligence tools. It said that on these facts, chats and documents generated through Claude had to be produced to prosecutors. (qz.com)