A coalition of news organizations accuses OpenAI of destroying roughly 20 million conversation logs that were supposed to be preserved as evidence in a landmark copyright case
When a court tells you to keep the receipts, you keep the receipts. OpenAI, according to a coalition of major news publishers, did not keep the receipts.
A group of plaintiffs led by The New York Times has filed a motion seeking sanctions against OpenAI, alleging the AI company deleted or failed to adequately preserve approximately 20 million ChatGPT conversation logs that a court had ordered it to retain. Those logs are central to the publishers’ claims that OpenAI scraped millions of copyrighted articles to train its flagship chatbot without permission, compensation, or so much as a courtesy email.
The evidence destruction allegations #
The sanctions request stems from what the plaintiffs describe as a direct violation of preservation directives issued by the court. In May 2025, US Magistrate Judge Wang ordered OpenAI to retain ChatGPT conversation logs as evidence in the case. The publishers now argue that OpenAI failed to comply with that order, and the missing data is precisely the kind of material that could demonstrate how ChatGPT ingests and reproduces copyrighted content.
The coalition includes not just The New York Times but also the New York Daily News and other affiliated publications. The original complaint was filed on December 27, 2023, making this litigation roughly two years old at the time of the sanctions request. In that time, the case has expanded and intensified considerably, with an amended complaint filed as recently as June 25, 2026, and summary judgment briefing concluding in April 2026.
Why conversation logs matter so much #
When users interact with ChatGPT, their prompts and the model’s responses create a real-time record of what the AI “knows” and how it presents information. If a user asks ChatGPT to summarize a New York Times article and the bot produces something suspiciously close to the original text, that log becomes direct evidence that OpenAI’s model has internalized copyrighted material in a way that competes with the original publisher.
Courts take evidence destruction seriously, especially when it happens after a preservation order has been issued. Sanctions can range from monetary penalties to adverse inference instructions, where a judge tells a jury it can assume the destroyed evidence would have been unfavorable to the party that destroyed it.
The broader copyright war over AI training #
OpenAI has generally argued that its use of publicly available text for training purposes falls under fair use, a legal doctrine that permits limited use of copyrighted material without permission under certain circumstances. The publishers counter that there’s nothing “limited” about ingesting millions of articles to build a commercial product that directly undermines the business model those articles support.
It’s worth noting that OpenAI has struck licensing deals with some publishers, including the Associated Press and several others. But the Times and its co-plaintiffs have not been interested in licensing arrangements, preferring instead to challenge the fundamental legality of what OpenAI did.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our