A US Judge Rules on AI Chats: No Privilege for Fraud Defendant
In a groundbreaking February ruling, Judge Jed Rakoff determined that Bradley Heppner, facing fraud charges, cannot claim privilege or work-product protection for chats he had with Anthropic's Claude AI about his legal exposure. This decision, considered the first of its kind in the US, sets a significant precedent regarding the confidentiality of AI conversations.
The Case Details:
Heppner, former chairman of bankrupt financial services company GWG Holdings and founder of Beneficent, was charged with securities and wire fraud in November 2025. Before retaining legal counsel, he used Claude to analyze his potential defenses and legal arguments, independently querying the AI about his situation. During an FBI search of his home, approximately 31 documents containing these AI conversations were seized, leading to a privilege dispute.
Key Rulings:
-
Attorney-Client Privilege: Rakoff ruled that Claude, being an AI without a law license or duty of loyalty, cannot form a privileged relationship with Heppner. The judge stated, "Heppner had disclosed it to a third party... which had no obligation of confidentiality."
-
Reasonable Expectation of Confidentiality: The judge examined Anthropic's terms of service and privacy policy, revealing that Heppner had consented to data collection, use for model training, and potential disclosure to third parties, including regulatory authorities, thereby negating any reasonable expectation of confidentiality.
-
Work-Product Protection: Since Heppner was not acting at the direction of his lawyers when chatting with Claude, and the documents did not reflect their attorney-client relationship, work-product protection did not apply.
Impact:
Over a dozen major law firms have since issued client advisories in response to this ruling, emphasizing the potential risks of using public AI chatbots for legal research or discussions. The case, United States v. Heppner, has sparked a wave of legal warnings across the country, urging individuals to proceed with caution when engaging with AI tools regarding sensitive legal matters.