Overview
- U.S. Magistrate Judge Ona Wang on Nov. 7 ordered OpenAI to produce 20 million de-identified consumer ChatGPT conversations by Nov. 14 or within seven days of completing de-identification.
- In a new filing, OpenAI asks the court to vacate or narrow the order, saying more than 99.99% of the logs have nothing to do with the case and that each log is a complete conversation, not a single prompt-output pair.
- OpenAI says it proposed privacy-preserving alternatives such as targeted searches and high-level classifications, which plaintiffs rejected, and it signals it may seek further review if the order stands.
- The New York Times and other news plaintiffs contend the conversations are needed for expert analysis of content reproduction, real-world use, and to rebut OpenAI’s claim that evidence was manufactured.
- OpenAI says the sample is a random draw of consumer chats from December 2022 through November 2024, excludes business customers, will be de-identified, and will be reviewed under a court protective order, with the judge citing a related Concord/Anthropic precedent.