OpenAI slams court order that lets NYT read 20 million complete user chats (arstechnica.com)

OpenAI asked a federal court to overturn an order forcing it to hand over 20 million complete ChatGPT conversations to The New York Times and other news organizations suing over alleged copyright infringement. OpenAI had earlier offered a 20M sample as a compromise to the NYT’s original request for 120M chats, but argues the court’s production order is overly broad because the logs are “complete conversations” — multiple prompt/output exchanges — and therefore much more likely to expose private information than isolated prompt-response snippets. The company says “more than 99.99%” of those chats are irrelevant to the case and has asked the court to vacate the order and require the plaintiffs to engage with OpenAI’s proposed process for identifying only truly relevant logs; it may also seek appellate review. The ruling has wider implications for AI/ML practice and data governance: compelled production of full-dialog logs could set a precedent for broad access to user interaction histories, risking user privacy, exposing personally identifiable information, and complicating how models are audited or defended in copyright suits. OpenAI highlighted current privacy leaks — some ChatGPT content has appeared in Google Search results and Search Console — and said it will build stronger safeguards, including client-side encryption for ChatGPT messages. The dispute underscores tensions between discovery in copyright litigation and the need for stringent technical protections around model training and conversational logs.