A federal judge has ruled that prosecutors can compel OpenAI to turn over ChatGPT account records belonging to Richard Kim, the former chief executive of cryptocurrency startup Zero Edge, as part of a fraud case alleging he diverted investor money into crypto trades and online gambling.

Prosecutors allege that Kim diverted about $3.8 million from a $4.3 million fundraising round for Zero Edge and that, following his arrest, he used ChatGPT to research his case, including his trial strategy. Court filings indicate that Kim may also have used the AI to generate numerous prompts related to misappropriating investor funds, cryptocurrency trading and gambling. Kim has pleaded not guilty to securities and wire fraud charges. 

US District Judge Lorna Schofield on Monday rejected the defense’s attempt to shield the chatbot data, establishing that AI chat logs can be treated as third-party digital evidence subject to a search warrant. The warrant seeks Kim’s OpenAI records from October 2023 through May 2026, including prompts, responses and account information.

The case is a reminder that conversations with AI chatbots can become part of a legal record. As more people use tools like ChatGPT for research and personal advice, courts are beginning to treat AI chats like other digital records, such as your emails, texts and search history. That also means using a chatbot for legal research doesn’t automatically make the conversation private or protected. 

Can a chatbot be protected by attorney-client privilege? 

Kim’s lawyers tried to block a search warrant, arguing that the chatbot data contains privileged information and research related to the case. According to the defense, those digital records should be protected, as they would expose Kim’s internal thoughts, defense tactics and trial strategy. 

Prosecutors countered that, for the attorney-client privilege to apply, a conversation must be confidential between a human and a licensed legal professional for the purpose of obtaining legal advice. An AI chatbot cannot be a lawyer.

Schofield’s ruling does not determine whether Kim’s ChatGPT records are protected by attorney-client privilege, but it does allow the warrant to proceed, meaning the defense cannot stop OpenAI from complying. Kim may still challenge specific records afterward.

The dispute adds to a growing legal question about whether conversations with tools like ChatGPT, Gemini or Claude can remain private when they are used for legal research.

This ruling aligns with a landmark decision from earlier this year in United States v. Heppner, in which another Manhattan judge found that a defendant’s exchanges with Anthropic’s Claude chatbot were not protected by attorney-client privilege or work-product protections.

In that case, US District Judge Jed Rakoff said that AI platforms are third-party data collectors, not legal counsel. He noted that the defendant used Claude without his attorneys’ direction and that the platform’s privacy terms weakened any claim that the chats were confidential. Rakoff said AI use directed by an attorney could be treated differently.

Future cases may draw sharper lines around when AI-assisted legal work can be protected, especially if a chatbot is used at an attorney’s direction. 

It’s not exactly a digital version of Miranda rights, but the warning here is similar: Anything you type can (and will) be used against you in a court of law.

(Disclosure: Ziff Davis, CNET’s parent company, filed a lawsuit against OpenAI in 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)