OpenAI blocks NYT from grabbing 20 million ChatGPT chats

Everything you have ever confessed to ChatGPT might soon be picked apart by a team of newspaper lawyers.

That sounds like an exaggeration, but it is currently playing out in a high-stakes court battle.

OpenAI Chief Information Security Officer Dane Stuckey just published an open pushback against The New York Times. As part of its copyright lawsuit against OpenAI, the publisher is asking a court to force OpenAI to hand over 20 million private user conversations. The Times claims it needs the logs to see if people used ChatGPT to bypass its paywall.

From 1.4 billion to 20 million

This is not the first time the publisher tried to look behind the curtain. According to OpenAI, The Times previously attempted to stop users from deleting their own chat histories, and at one point demanded a staggering 1.4 billion private conversations. OpenAI fought both demands off.

This latest 20 million target is a random sample of consumer conversations logged between December 2022 and November 2024. The Times argued in court that the request is reasonable because another AI company previously agreed to turn over 5 million user chats in an unrelated lawsuit.

OpenAI offered compromises, such as running targeted searches to only pull chats containing text from Times articles, or providing high-level classification data. The Times rejected both.

If you use ChatGPT through work or school, you are clear for now. The request only covers consumer accounts. ChatGPT Enterprise, Edu, Business, and API users are entirely excluded.

Rolling out the encryption shield

OpenAI is appealing the demand, but it is also preparing contingency defenses. If forced to comply, OpenAI plans to scrub all personal identifying information, like names and passwords, before letting anyone look.

Even then, access would be restricted to outside attorneys and technical consultants in a tightly locked-down review environment.

More importantly, OpenAI is accelerating its security roadmap to include client-side encryption. That means messages will be encrypted before leaving your device, making private chats inaccessible to third parties and even to OpenAI itself.

Why it matters

We think this battle marks a dangerous turning point where copyright disputes cross the line into broad privacy dragnets.

For everyday users, it is a wake-up call. The prompts you type into conversational AI do not just evaporate into the ether; they sit on corporate servers subject to legal discovery. While enterprise tiers enjoy contractual isolation, consumer accounts remain vulnerable. Until client-side encryption is fully deployed, users should think twice before typing highly sensitive personal details into any AI prompt.

For the industry, The Times' strategy creates a chilling precedent. If plaintiffs can routinely demand millions of unrelated user logs on fishing expeditions, every frontier AI lab will face massive pressure to deploy end-to-end encryption immediately.

OpenAI played its cards brilliantly here. By taking the dispute public, it flipped the narrative from corporate copyright infringement to defending 800 million weekly users against an invasive publisher.