An executive who received a grand jury subpoena and believed he was the target of a federal investigation turned to Claude, a prominent Artificial Intelligence program, and without direction from a lawyer that he do so, “prepared reports that outlined defense strategy, that outlined what he might argue with respect to the facts and the law that we anticipated that the government might be charging.” He later gave them to his lawyer.
After he was arrested, the FBI searched his home, seizing documents and electronic devices which included documents he prepared using Claude.
The problem? None of those documents were confidential or protected by the attorney client privilege. In United States v. Hepner, a New York federal District Court held the documents were not protected by the attorney client privilege because they were not communications with his lawyer — Claude, a third party AI platform, “is not an attorney.” Importantly, Claude’s written privacy policy to which users consent, makes clear that its data is not confidential. Claude’s owner, Anthropic, collects users’ inputs and its outputs, uses that data to “train” Claude, and “reserves the right to disclose such data to a host of ‘third parties,’ including governmental regulatory authorities.”
Lawyers have long been warned that use of public AI platforms can breach attorney client privilege. In this case, documents created by a non lawyer using AI to analyze defenses to the charges against him were not confidential and therefore fully available to the government.
This decision suggests it is important that AI platforms not be used unless in consultation with counsel because information transmitted and obtained is likely not confidential, meaning it is potentially discoverable in criminal and civil matters.