According to Ars Technica, OpenAI is fighting a federal court order that would force the company to hand over 20 million private ChatGPT conversations to The New York Times and other news plaintiffs by November 14, 2025. The order came from US Magistrate Judge Ona Wang on November 7, who sided with the NYT despite OpenAI’s privacy concerns. OpenAI had previously offered the 20 million chat sample as a counter to the NYT’s original demand for 120 million conversations. The company argues that more than 99.99% of these conversations have nothing to do with the copyright infringement case and that disclosing complete chat logs exposes far more private information than individual prompt-output pairs. OpenAI says the NYT wants these conversations to find examples of users trying to bypass their paywall using ChatGPT.
The privacy precedent that worries everyone
Here’s the thing that should concern every ChatGPT user: OpenAI is arguing this sets a dangerous precedent where anyone suing an AI company could demand tens of millions of private conversations without proving relevance first. They’re basically saying your ChatGPT chats should be treated with the same privacy protection as your Gmail emails. And they’ve got a point – would any court let plaintiffs dig through 20 million private emails just because they’re suing Google?
The company filed a detailed legal argument explaining why complete conversation logs are fundamentally different from individual prompt-response pairs. Each of those 20 million logs represents an entire exchange between a user and ChatGPT, which could mean up to 80 million individual prompt-output pairs. That’s like someone eavesdropping on your entire therapy session versus just hearing one question and answer.
The de-identification debate
Judge Wang ruled that user privacy would be protected by existing legal safeguards and OpenAI’s “exhaustive de-identification” process. But OpenAI pushes back hard on this, saying de-identification doesn’t remove information that’s non-identifying but still private. Think about a journalist using ChatGPT to help draft a sensitive news article – even if you remove their name, the content itself could reveal confidential sources or unpublished stories.
OpenAI claims they never got a chance to argue why the Concord Music case that the judge referenced doesn’t apply here. That case involved 5 million records, but they were just single prompt-output pairs, not complete multi-exchange conversations. There’s a huge difference between seeing one question and answer versus reading an entire back-and-forth discussion.
What’s really at stake here
The NYT isn’t just being difficult for the sake of it. They filed their own motion arguing that they need the actual model outputs to understand how “real world” users interact with ChatGPT, especially around news content and paywall bypassing. They say OpenAI’s offer to run searches on their behalf is inadequate for proper expert analysis.
But let’s be real – this feels like fishing expedition territory. When over 99.99% of the conversations have nothing to do with the case, why should millions of users have their private chats exposed? OpenAI says they offered targeted searches and high-level data classification, but the Times rejected those approaches and demanded the entire dataset on a hard drive.
The bigger privacy picture
This legal fight comes amid growing concerns about ChatGPT privacy overall. The company acknowledges that ChatGPT conversations have appeared in Google search results and developer tools. In response, they’re promising “advanced security features” including client-side encryption for messages.
OpenAI posted a message to users assuring them that the chats are stored in a secure system under legal hold, meaning they can’t be accessed for purposes other than meeting legal obligations. They’ve committed to fighting any attempts to make user conversations public outside the court process.
So where does this leave us? We’re watching a fundamental battle over digital privacy play out in real time. Does using an AI chatbot mean surrendering your expectation of privacy in legal disputes? The outcome could shape how all AI companies handle user data for years to come. And honestly, it makes you think twice about what you share with any AI assistant.
