A federal judge just signed off on searching Richard Kim’s OpenAI account. He used to run a crypto startup called Zero Edge. Prosecutors say he stole $3.8 million. They think he chatted with an AI about it.
The warrant
US District Judge Lorna Schofield made the call. The defense tried to block it, claiming the chat logs were privileged. They lost. The judge decided that AI chats count as third-party evidence. Just like emails. Or search history.
The warrant covers records from October 2023 to May 2026. It includes prompts, answers, and account details. Kim says he used ChatGPT to study his trial strategy after getting arrested. Prosecutors aren’t buying that. Or maybe they just don’t care if the advice came from a bot or a book.
Not a lawyer
Kim’s lawyers argued that the chats contained internal thoughts. They said it would expose defense tactics. That’s too risky to share with a judge, they claimed.
Prosecutors hit back hard. Attorney-client privilege is specific. It requires a human. A licensed one. An algorithm isn’t a lawyer.
The ruling doesn’t decide if the chats are privileged, but it lets the search happen first. Kim can still fight over specific files later.
This isn’t new ground. Remember United States v. Heppner earlier this year? Another Manhattan judge said the same thing. In that case, the defendant chatted with Claude. Judge Jed Rakoff shut it down. He said AI platforms collect data. They are not counsel.
Rakoff pointed out that the guy in that case didn’t use Claude under a lawyer’s direct orders. Also, the privacy policy basically said, we own your data. If an attorney tells you to use it? Maybe that changes. We’ll see.
The trap
Here’s the kicker. People use ChatGPT for advice now. It feels private. It’s just you and a screen. It isn’t.
Whatever you type goes into a system owned by someone else. You don’t have control there. Using a chatbot for legal research isn’t a magic shield. It doesn’t make your thoughts invisible.
Maybe this is the new digital Miranda warning. Not exactly the same, sure. But the effect is clear. Anything you type? It can be used against you. And it probably will be.
Who checks their terms of service before confessing to a crime to a robot?
(Disclosure: CNET’s parent, Ziff Davis, sued OpenAI in 2021. They claimed the AI stole their writing style.)
