A court rules a defendant's Claude conversations are not privileged
In shortA defendant typed what his lawyers had told him into Claude. A federal judge ruled the documents it produced were not protected.
What happened
Benjamin Heppner, indicted in October 2025 on securities and wire fraud charges, used Anthropic's Claude without any suggestion from his lawyers. He entered information he had learned from them and generated reports outlining defence strategy. The FBI seized about 31 such documents at his arrest.
On 10 February 2026 Judge Jed Rakoff of the Southern District of New York ruled from the bench that the documents were not protected, and gave written reasons on 17 February. Claude is not an attorney, its privacy policy allows inputs and outputs to be disclosed to third parties, and the documents were not prepared at counsel's direction, so they were not work product either.
Why it worked
Nothing was breached. The information left the privileged relationship when it was typed into a tool whose terms allow disclosure. The court did not decide how a tool used at counsel's direction would be treated.
What EndLayer shows or does
- Available now
The Shadow AI queue lists the AI websites reached from enrolled devices that nobody has ruled on, so a firm knows which tools its people use beyond the ones it chose. Roadmap →
- Available now
An AI site on your blocklist is blocked by your MDM, and the agent checks that the block is in force. Roadmap →
- Available now
The browser extension checks what a person types into an AI site's composer before it is sent, with the built-in detectors and the patterns you write. Roadmap →
- Not covered
Whether a document is privileged. That is a legal question, and no detector recognises privileged material as a category. A device that is not enrolled is not seen.