OpenAI District Court Orders That Allow The New York Times To Access 20 Million ChatGPT User Conversations

JAKARTA - OpenAI is asking the court to cancel a decision requiring the company to hand over 20 million ChatGPT users' conversations to The New York Times (NYT) and a number of other media that sued him for alleged copyright infringement.

Although OpenAI previously offered 20 million conversations as an alternative to the NYT request which reached 120 million logs, the company considered the court order to be too broad and risky to user privacy.

The requested "Log includes a complete conversation each log contains a full sequence of several prompt-output pairs between users and ChatGPT," OpenAI wrote in a file in the US District Court for the Southern District of New York. The disclosure of the log is much more likely to reveal personal information, as is the case with listening to the whole conversation compared to just a five-second snippet.

OpenAI confirmed that more than 99.99% of the conversations were irrelevant to the proposed cases. Therefore, the company asked the court to cancel the order and ordered the plaintiffs to respond to OpenAI's more limited proposal.

The company also announced on its official website that: The New York Times demands that we submit 20 million of your private conversations with ChatGPT to find examples of users trying to get past their news paywalls.

In court documents, OpenAI asserts that the production of conversation logs should be limited only to relevant data. "OpenAI is not aware of any case where the court ordered the production of personal information on this scale," the company wrote.

This creates a dangerous precedent: anyone suing an AI company could sue millions of users' conversations without relevance. The court has never allowed those who demand Google to search through the personal emails of millions of Gmail users for no relevant reason and the same should apply to the generative AI.

However, Judge Magistrat Ona Wang on November 7 decided that OpenAI must submit 20 million log conversations that have been anonymized to the plaintiffs no later than November 14, 2025, or seven days after the de-identification process is complete.

According to the judge, user privacy is sufficient to be protected by a legal protection order and anonymization process carried out by OpenAI itself.

OpenAI denied the assessment, citing that the process of anonymity does not always erase non-identificational personal information, such as the example of a Washington Post reporter who uses ChatGPT to write news articles.

NYT also responded to OpenAI's statement. In its official statement, the media said: "This lawsuit aims to hold OpenAI and Microsoft accountable for stealing millions of copyrighted works to create products that compete directly with The Times."

"OpenAI deliberately misleads the public by scaring users, even though there is no privacy risk whatsoever. The court only asks for examples of conversations that have been anonymized by OpenAI itself under legal protection".

OpenAI explained that the 20 million conversations were taken randomly from December 2022 to November 2024, and did not cover business users.

OpenAI claims to have offered a number of privacy options to NYT, such as:

The search is limited to log which has the potential to contain the text of the NYT article,

as well as the provision of aggregate data on how users take advantage of ChatGPT.

However, all of these options were rejected by the NYT.

The conversation data is now stored in a protected system under a legal hold', meaning it cannot be accessed or used except for legal purposes. NYT is also legally bound not to publish the data outside the court process.

NYT accused OpenAI of refusing to comply with previous deals by not submitting an example of the requested model output. The media argued that direct access to conversation logs was needed so that experts could analyze how models work, real user interactions, and how often ChatGPT generates content from the news.

Meanwhile, OpenAI confirmed that the request had shifted from log' to NYT content to a request for a full of 20 million complete conversations.

OpenAI also rejected Judge Wang's comparison with the Concord Music Group v. Anthropic PBC case, as the case involved 5 million single prompt-output pairs, not full conversations like ChatGPT.

"In the Concord case, the only data requested is a single pair, not a complete series of conversations that can cover up to 80 million interactions," OpenAI explained.

OpenAI plans to appeal this ruling and warns that the decision could set a dangerous precedent for future user privacy.

The company also announced that it will develop advanced security features, including client-side encryption to protect ChatGPT users' messages from accessing third parties.