Chats stored under legal hold

The 20 million chats consist of a random sampling of ChatGPT conversations from December 2022 to November 2024 and do not include chats of business customers, OpenAI said in the message on its website.

“We presented several privacy-preserving options to The Times, including targeted searches over the sample (e.g., to search for chats that might include text from a New York Times article so they only receive the conversations relevant to their claims), as well as high-level data classifying how ChatGPT was used in the sample. These were rejected by The Times,” OpenAI said.

The chats are stored in a secure system that is “protected under legal hold, meaning it can’t be accessed or used for purposes other than meeting legal obligations,” OpenAI said. The NYT “would be legally obligated at this time to not make any data public outside the court process,” and OpenAI said it will fight any attempts to make the user conversations public.

A NYT filing on October 30 accused OpenAI of defying prior agreements “by refusing to produce even a small sample of the billions of model outputs that its conduct has put in issue in this case.” The filing continued:

Immediate production of the output log sample is essential to stay on track for the February 26, 2026, discovery deadline. OpenAI’s proposal to run searches on this small subset of its model outputs on Plaintiffs’ behalf is as inefficient as it is inadequate to allow Plaintiffs to fairly analyze how “real world” users interact with a core product at the center of this litigation. Plaintiffs cannot reasonably conduct expert analyses about how OpenAI’s models function in its core consumer-facing product, how retrieval augmented generation (“RAG”) functions to deliver news content, how consumers interact with that product, and the frequency of hallucinations without access to the model outputs themselves.

OpenAI said the NYT’s discovery requests were initially limited to logs “related to Times content” and that it has “been working to satisfy those requests by sampling conversation logs. Towards the end of that process, News Plaintiffs filed a motion with a new demand: that instead of finding and producing logs that are ‘related to Times content,’ OpenAI should hand over the entire 20 million-log sample ‘via hard drive.’”