Chats may be discoverable in court

U.S. lawyers are warning clients that chatbot conversations can end up in court after a federal judge in New York refused to treat one defendant’s artificial intelligence chats as privileged. (usnews.com) The ruling came in *United States v. Bradley Heppner*, a Southern District of New York case decided on February 17, 2026. The court held that a criminal defendant’s exchanges with a publicly available generative artificial intelligence platform were not protected by attorney-client privilege or work-product doctrine. (perkinscoie.com) Reuters reported on April 15 that more than a dozen large U.S. law firms have since sent client alerts, updated website guidance, or added contract language warning that chats with tools such as ChatGPT and Claude could be sought by prosecutors or civil litigants. (usnews.com) Attorney-client privilege is the rule that usually keeps private communications between a lawyer and client out of evidence. The Heppner court said that protection did not apply because a chatbot is not a lawyer, the exchanges were not confidential in the legal sense, and the materials were created without counsel’s direction. (adamsandreese.com) The facts in the case were unusually concrete. According to court summaries by Perkins Coie and Adams and Reese, Heppner used Anthropic’s Claude in 2025, after learning he was a target of a criminal investigation, to prepare reports about possible defense strategy; Federal Bureau of Investigation agents later seized about 31 documents tied to those exchanges. (perkinscoie.com) (adamsandreese.com) The court also focused on platform terms and privacy disclosures. Perkins Coie said the judge noted that the service collected user inputs and outputs, used them to train the model, and reserved the right to disclose data to third parties, including government authorities. (perkinscoie.com) That analysis reaches beyond one fraud case because discovery rules already cover electronically stored information such as messages, drafts, and search histories. K&L Gates wrote in 2025 that prompts and outputs from tools like ChatGPT may qualify as discoverable electronically stored information under federal and state rules. (khflaw.com) Law firms are responding by telling clients not to paste legal advice, strategy notes, or privileged communications into public chatbots. Reuters said Sher Tremonte recently added contract language warning that sharing a lawyer’s advice with a chatbot could waive attorney-client privilege. (usnews.com) The risk is not identical across products. OpenAI says business data in ChatGPT Enterprise, ChatGPT Business, ChatGPT Edu, ChatGPT for Healthcare, ChatGPT for Teachers, and its application programming interface platform remains under customer control and is not used to train models by default. (openai.com 1) (openai.com 2) Consumer settings offer some limits, but not a courtroom shield. OpenAI says Temporary Chat conversations do not appear in history and are not used to train models, while help-center documentation says retention and backup periods can still apply and enterprise compliance tools can still access temporary chats. (openai.com) (help.openai.com 1) (help.openai.com 2) A separate court fight has already sharpened those retention concerns for ChatGPT users. The Bar Association of San Francisco wrote in August 2025 that a court order in *The New York Times v. OpenAI* required OpenAI to preserve chat logs, including some deleted chats, for many consumer accounts, while enterprise accounts had stronger protections. (sfbar.org) The practical line from lawyers is narrower than a blanket ban on artificial intelligence: use secure products with contractual controls for low-risk work, and keep sensitive facts, admissions, and legal strategy inside channels that are actually protected. The Heppner ruling turned that advice from a caution into litigation practice. (perkinscoie.com) (usnews.com)