AI chats discoverable
U.S. lawyers are warning that conversations with AI chatbots can be treated as discoverable evidence in court, creating new legal risk for companies and users. Reuters reports a recent ruling has pushed firms to rethink whether chat logs are private or subject to disclosure in litigation (reuters.com). At the same time, privacy audits and cases show platforms are under scrutiny for handing data to law enforcement, possible state-law violations, and user access disputes — from an EFF complaint about Google and ICE to a Dutch appeal ordering X to give a researcher his personal data (eff.org) (nltimes.nl).
A federal judge in Manhattan has ruled that some chatbot conversations can be collected in court, and lawyers are warning clients to treat them like ordinary records, not private confidences. (money.usnews.com) (crowell.com) The ruling came from United States District Judge Jed Rakoff in the Southern District of New York in the criminal case against Bradley Heppner. Rakoff ruled from the bench on February 10, 2026, and issued a written opinion on February 17 saying 31 Claude-generated documents were not protected by attorney-client privilege or the work-product doctrine. (crowell.com) (debevoise.com) Rakoff said a public artificial intelligence tool is a third party, not a lawyer, and that sharing legal analysis with it can destroy confidentiality. Reuters reported on April 15 that more than a dozen large United States law firms have since issued client warnings about using tools such as Claude and ChatGPT for sensitive matters. (crowell.com) (money.usnews.com) The legal issue is discovery, the process that lets prosecutors or civil litigants demand relevant documents and messages before trial. If a chat log sits on a company server or in a user account without a recognized privilege, it can become evidence like an email, text, or spreadsheet. (money.usnews.com) (harvardlawreview.org) Law firms are responding by steering clients toward closed enterprise systems, tighter retention settings, and prompts that avoid pasting in legal advice or confidential facts. O’Melveny and Crowell & Moring are among the firms that have published alerts saying the risk turns on which tool is used, what data goes in, and what contractual protections exist. (omm.com) (crowell.com) The privacy fight goes beyond courtroom privilege. On April 14, the Electronic Frontier Foundation said it asked the California and New York attorneys general to investigate Google after the company gave Amandla Thomas-Johnson’s account data to Immigration and Customs Enforcement in May 2025 without advance notice, despite a long-running policy of notifying users before disclosure. (eff.org) Thomas-Johnson wrote that Immigration and Customs Enforcement sent Google an administrative subpoena in April 2025 after he had attended a pro-Palestinian protest at Cornell University in September 2024. Google said in a statement quoted by Reuters that it has “long supported notice to users” and reviews every government demand for legal validity, while also saying it may delay notice when required by law or in emergencies. (eff.org) (money.usnews.com) Courts are also pressing platforms to hand users their own data. On April 14, the Amsterdam Court of Appeal ruled that X must give privacy researcher Danny Mekić access to personal data the company holds on him after his account was hidden from other users in October 2023. (nltimes.nl) The appeal court left some limits in place, saying X did not have to reveal employee identities or precise timestamps that could expose how its moderation systems work. But it upheld the core point that users can force disclosure of at least some platform-held records about how their accounts were handled. (nltimes.nl) The result is a narrower idea of what stays private when people type into software run by someone else. In court, at the border, or in a data-access case, the chat box is starting to look less like a diary and more like a file cabinet. (money.usnews.com) (eff.org) (nltimes.nl)
