Courts may use your AI chats
U.S. lawyers are warning that a recent ruling makes it possible for chatbot conversations to be used in legal proceedings, meaning chats should not be treated as privileged or private by default. The alert has immediate implications for how employees and counsel record sensitive information in AI tools. (reuters.com)
A federal judge in New York ruled in February that chats with a public artificial intelligence bot can be examined by prosecutors in court. (law.com) The case is *United States v. Bradley Heppner*, in the Southern District of New York. Judge Jed S. Rakoff issued a bench ruling on February 10, 2026, and a written opinion on February 17, 2026, after the Federal Bureau of Investigation seized about 31 documents reflecting Heppner’s exchanges with Anthropic’s Claude. (perkinscoie.com) Heppner had used Claude after receiving a grand jury subpoena and after learning he was a target of a criminal investigation, according to court summaries. His lawyers said he used the bot to prepare reports on defense strategy and possible arguments on the facts and the law. (harvardlawreview.org) Attorney-client privilege is the rule that can keep private communications between a lawyer and client out of court. Work product is a separate rule that can protect documents prepared for litigation, especially a lawyer’s mental impressions and strategy. (law.cornell.edu) Rakoff said the Claude exchanges did not qualify for either protection. The court said Claude was not a lawyer, the chats were shared with a third party, and the defendant had no reasonable expectation the material would stay confidential. (perkinscoie.com) The opinion turned in part on the product’s own terms. Court summaries said Anthropic’s policy disclosed that it collected user inputs and outputs, used data to train Claude, and could disclose data to third parties, including government authorities. (debevoise.com) Lawyers have spent the past two months warning clients and employees not to paste sensitive facts, legal theories, witness accounts, or draft arguments into consumer chatbots. Reuters reported on April 15 that United States lawyers are now telling clients not to treat ChatGPT, Claude, and similar tools as private confidants when criminal exposure or civil liability is at stake. (reuters.com) Some lawyers say commentary around the ruling has gone too far. Venable said the decision does not mean every use of artificial intelligence automatically destroys privilege, and argued future cases could turn on facts such as lawyer direction, confidentiality terms, and whether the tool operates inside a protected enterprise environment. (venable.com) That leaves a narrower but immediate lesson from *Heppner*: if a person types litigation strategy into a public chatbot before sending it to a lawyer, a court may treat that material as something the other side can read. (orrick.com)
