A federal judge in the Southern District of New York has ordered OpenAI to surrender 20 million anonymized ChatGPT conversation logs to a coalition of news organizations, marking a significant development in ongoing copyright litigation.
District Judge Sidney H. Stein upheld Magistrate Judge Ona T. Wang’s earlier ruling, rejecting OpenAI’s attempts to limit disclosure on user privacy grounds.
The Discovery Dispute
The order stems from a consolidated set of 16 copyright cases filed by major content creators, including The New York Times and the Chicago Tribune.
The plaintiffs argue that OpenAI unlawfully used their copyrighted works to train its artificial intelligence models.
The discovery battle began in July, when news organizations initially requested 120 million conversation logs to analyze how the AI generates responses from their content.
As part of the compromise, OpenAI agreed to produce 20 million logs, representing approximately 0.5% of its preserved data.
However, the agreement broke down in October when OpenAI refused to provide the complete random sample and instead offered a limited set of search results that specifically implicated the plaintiffs’ works.
Magistrate Judge Wang ruled in favor of the publishers in November and denied OpenAI’s request for reconsideration in December, prompting the company to appeal to the District Judge.
In its appeal, OpenAI argued that the court failed to weigh user privacy risks adequately and should have mandated the “least burdensome” discovery method.
The company cited a Second Circuit securities precedent where discovery of SEC call recordings was blocked due to wiretapping concerns.
Judge Stein dismissed these arguments on Monday, distinguishing the current case from the securities precedent.
He noted that, unlike secret wiretaps, ChatGPT users voluntarily submit their queries to the platform, and OpenAI’s ownership of these logs is undisputed.
The court affirmed that Judge Wang had balanced correctly the potential privacy intrusions against the material’s critical relevance to the litigation, stating there is no binding case law requiring courts to default to the least burdensome discovery option.
This ruling marks a pivotal moment in the multidistrict litigation and could set the standard for how intellectual property law applies to generative AI.
The outcome of these cases is expected to determine the legal boundaries for how AI companies ingest and utilize copyrighted material for model training.
Follow us on Google News , LinkedIn and X to Get More Instant Updates. Set Cyber Press as a Preferred Source in Google.