Richard Kim allegedly used ChatGPT to research his own defense after prosecutors say he diverted $3.8 million of investor money. A federal judge just ruled those prompts are producible under a search warrant.
He allegedly turned to ChatGPT to research his own criminal defense. Now those prompts are evidence the government can read.
A federal judge in Manhattan ruled Monday that prosecutors can compel OpenAI to turn over the ChatGPT account records of Richard Kim, the former CEO of cryptocurrency startup Zero Edge, rejecting defense arguments that the chatbot conversations should be shielded from disclosure. The decision is the first publicly reported ruling to treat ChatGPT account data as third-party digital evidence producible under a standard search warrant, putting AI chats in the same legal bucket as email and cloud storage rather than under any form of attorney-client privilege.
Kim has pleaded not guilty to charges of securities and wire fraud. Prosecutors at the U.S. Attorney's Office for the Southern District of New York allege he diverted roughly $3.8 million of a $4.3 million Zero Edge fundraising round into cryptocurrency trades and online gambling. After his arrest, according to filings summarized by CNET and the New York Law Journal, Kim used ChatGPT to research his own case and trial strategy, and prosecutors also cite prompts about misappropriating investor funds and gambling.
U.S. District Judge Lorna Schofield rejected the defense's bid to treat the chatbot data as private. The warrant, reported by the New York Law Journal, seeks OpenAI records from October 2023 through May 2026, including prompts, responses, and account information.
For anyone who has used an AI assistant to think through a sensitive question, the practical takeaway is blunt: conversations with a third-party AI vendor carry no special legal shield. They are producible under the same warrant authority that already covers Gmail and search history.
That is not a novel Fourth Amendment conclusion. The third-party doctrine has long held that information voluntarily shared with a service provider belongs to the records of that provider, not the user. What is new is the category of record: a conversation that can feel private because an AI assistant replies in natural language and at conversational speed, often in a tone calibrated to sound supportive. Courts have not yet decided whether specific exchanges in a ChatGPT history could carry a separate privilege, such as when a prompt is paired with a privileged attachment or a user is communicating through the chatbot with a licensed attorney. Schofield's ruling only resolves whether the warrant can proceed; record-by-record privilege fights remain live.
The Kim case sits inside a wider conversation about AI in legal practice. A March 2026 Harvard Law Review blog post on United States v. Heppner traces how courts have begun treating AI-generated outputs as evidence, and a separate National Law Review piece catalogs early rulings that ask whether AI tools themselves may be edging into the unauthorized practice of law. Neither precedent governs Kim, but both signal that the bench is no longer treating AI outputs as legally inert.
What to watch next: whether Kim's defense team pursues a privilege challenge at the production stage, whether other federal magistrates cite Schofield's reasoning in the months ahead, and whether the privilege question reaches a published appellate opinion. The doctrine being built here is small in the Kim docket and large for everyone else.