News publishers seek sanctions against OpenAI over ChatGPT logs

News organizations suing OpenAI have asked a court to impose serious sanctions, accusing the company of concealing evidence tied to whether ChatGPT reproduced copyrighted journalism. The dispute matters because ChatGPT output logs could help determine whether OpenAI’s use of news content qualifies as fair use or shows copyright infringement, according to the publishers’ filing.

In a sanctions motion filed Thursday, plaintiffs led by The New York Times alleged that OpenAI misled them and the court for two years about its ability to search ChatGPT logs. The publishers said an April deposition of OpenAI privacy engineer Vincent Monaco showed the company had already searched large anonymized log samples before the litigation began.

OpenAI disputed the accusations. A company spokesperson said the motion was part of an effort by the Times to obtain more user logs and said OpenAI would keep defending user privacy and fair use.

Dispute over hidden log samples

The publishers alleged that OpenAI had two de-identified datasets containing 10 million and 78 million logs, but did not disclose them during discovery. According to the filing, OpenAI had also searched those datasets for Times content while researching a filter meant to block the reproduction of copyrighted material.

The plaintiffs said OpenAI’s position changed depending on its interests. They argued that the company claimed broad searches were infeasible, costly and privacy-invasive when publishers sought access, while using similar searches internally.

The filing said the concealed samples could have been made available early in the case, giving the publishers more time to review potential evidence. The publishers argued that OpenAI’s conduct increased discovery costs, delayed the case and burdened the court.

Redactions and preservation fight

The plaintiffs said OpenAI instead gave them access to a “sandbox” containing a 20 million-log sample, after the publishers had sought 120 million news-related logs. According to the sanctions motion, OpenAI used AI tools to make 19 billion redactions in that sample, and the court later found the resulting dataset unusable.

OpenAI later removed some redactions, the publishers said, but they alleged that remaining redactions to domains, names and other fields continued to limit their searches. The plaintiffs also said OpenAI had a de-identified 78 million-conversation sample throughout that period.

The filing also accused OpenAI of deleting parts of the 20 million-log sample and of deleting or compressing billions of logs that should have been preserved. According to the publishers, Monaco testified that OpenAI considered complying with a court preservation order requiring retention of chats, decided it would be difficult, and took no steps to do so.

OpenAI has previously opposed broad log production on privacy grounds. Its spokesperson said the Times was making “blatantly false allegations” as its case weakened and said the company would continue to defend users’ privacy.

Publishers ask for severe penalties

The publishers are asking the court to bar OpenAI from relying on the 20 million-log sample. They also want the court to find that withheld output logs contained substantial reproductions of their copyrighted material and to prevent OpenAI from arguing otherwise, according to the filing.

The plaintiffs further asked that jurors be told OpenAI deleted billions of logs. They argued that lesser penalties would not address what they described as knowing and intentional misconduct.

The Times has rejected OpenAI’s claim that the case is weakening. Times spokesperson Graham James previously said the plaintiffs had streamlined the suit and strengthened it by adding claims against Microsoft, while keeping the core allegation that Microsoft and OpenAI used Times journalism without permission to compete with the publisher’s products and enrich themselves.

This story draws on original reporting from Ars Technica.