OpenAI: NYT Demands 20 Million Private ChatGPT Conversations

OpenAI is publicly pushing back against what it calls a massive overreach by The New York Times, which is demanding access to 20 million private ChatGPT conversations as part of an ongoing copyright lawsuit. In a blog post authored by Chief Information Security Officer Dane Stuckey, the company framed the legal maneuver as a direct threat to the privacy of its 800 million weekly users, arguing the conversations are being treated as collateral in a dispute over online content access. The Times’ lawyers, OpenAI claims, are looking for evidence that users might have used ChatGPT to circumvent the publication’s paywall.

This isn’t the first volley in this particular discovery fight. OpenAI says the Times initially sought to strip users of their ability to delete chat history, a demand the AI company successfully fought. A subsequent request, according to Stuckey, ballooned to 1.4 billion conversations before being narrowed—though still unacceptably broad in OpenAI’s view—to the current 20 million. The company says it even offered privacy-preserving alternatives, such as targeted searches for chats containing NYT article text, but those were rejected. Stuckey didn’t mince words, stating the demand disregards long-standing privacy protections and would force OpenAI to hand highly personal data to the Times’ lawyers and paid consultants.

To counter the legal pressure, OpenAI is accelerating its security roadmap. The most significant technical commitment is the development of client-side encryption for ChatGPT messages, a feature that would theoretically make conversations inaccessible even to OpenAI itself. Stuckey also detailed plans for fully automated safety detection systems, reserving human review for only the most critical edge cases involving threats to life or cybersecurity. These features, he noted, are in active development with more details promised in the near future.

Behind the legal wrangling is a stark warning from OpenAI about the precedent this would set. The company noted that the Times’ lawyers justified their request partly because another AI firm handed over 5 million private chats in a separate case—a comparison OpenAI firmly rejects. While the Times is legally obligated to keep the data private during court proceedings, OpenAI vows to fight any attempt to make the conversations public. For a company that acknowledges it is one of the most targeted organizations in the world by state-sponsored hackers, being forced to create a pipeline of private user data for a civil lawsuit represents a fundamental conflict with its security mandate. The core tension remains unresolved: how do you prove a model wasn’t trained on copyrighted material without exposing the private lives of the people who used it?