The New York Times and other news publishers asked a federal court Thursday to sanction OpenAI, accusing the company of concealing ChatGPT output logs that could bear directly on whether the chatbot reproduced copyrighted journalism.
The fight matters because both sides have treated the logs as central evidence. The publishers say user prompts and ChatGPT responses could show people using the service to get around paywalls or obtain near-verbatim copies of articles. OpenAI says its systems are protected by fair use and has argued that broad access to chat logs would invade user privacy.
In a sanctions motion, the publishers claimed OpenAI spent about two years telling the court that searching large sets of ChatGPT logs would be technically difficult, costly, and privacy-invasive, while the company had already searched large de-identified datasets before the case began. The filing says that claim emerged during an April deposition of OpenAI privacy engineer Vincent Monaco, whom the publishers describe as unprepared.
According to the motion, Monaco testified that OpenAI had two large de-identified samples, one containing 10 million logs and another containing 78 million logs. The publishers say OpenAI never disclosed those datasets during discovery, even though they could have been used much earlier in the case.
The motion also says OpenAI had already searched those samples for New York Times material while researching a filter intended to block outputs that reproduced copyrighted content. The publishers argue that undercuts OpenAI’s position that comparable searches were infeasible or too burdensome when requested in litigation.
The log fight
The publishers originally sought access to a much larger set of news-related logs. Instead, according to the motion, they spent months working in a controlled “sandbox” with a 20 million-log sample that OpenAI heavily redacted.
The filing says OpenAI used AI systems to apply 19 billion redactions to that smaller dataset. The court later found the sample unusable, according to the publishers. OpenAI eventually removed some redactions, but the motion says many remained, including fields involving publisher domains and names, which limited the searches the plaintiffs could run.
The publishers also accuse OpenAI of deleting or compressing billions of logs that should have been preserved under a court order requiring retention of chats. The motion says Monaco testified that OpenAI considered compliance difficult and did not take steps to preserve all covered material.
OpenAI disputed the allegations. A company spokesperson said the Times was trying to obtain more user logs after weakening its case, and called the publishers’ accusations false. The spokesperson said OpenAI would keep defending user privacy and fair use.
Graham James, a spokesperson for the Times, previously disputed OpenAI’s characterization after the publishers dropped some claims. James said the case had been streamlined and strengthened by adding claims against Microsoft, and said the core allegation remained that Microsoft and OpenAI used millions of Times works without permission to compete with the publisher.
Ian Crosby, lead counsel for the Times, told Ars Technica that OpenAI had misled the publishers, the public, and the court by claiming searches for copied news content were infeasible while hiding that such searches had already been done.
What the publishers want
The publishers are asking the court to impose severe penalties. Their requested sanctions include barring OpenAI from relying on the 20 million-log sample, deeming withheld logs to contain substantial reproduction of the publishers’ copyrighted material, and instructing the jury that OpenAI deleted billions of logs.
If granted, those sanctions could damage OpenAI’s fair-use defense. The copyright case is expected to turn in part on whether the publishers can show market harm, including whether ChatGPT substituted for news products by reproducing articles. OpenAI’s position is that its technology is lawful and transformative, but the discovery fight has now become its own problem for the court to sort out.
This story draws on original reporting from Ars Technica.
