A remedy, not a recall notice
The sharpest line in the complaint is not the $250 million. It is the request that OpenAI destroy every large language model and training dataset that incorporates the publishers' work.
That does not mean a judge is about to order ChatGPT deleted. It is a remedy the plaintiffs want, and OpenAI still has the opportunity to contest the claims, the evidence and the legal theory. The court has not found infringement. The demand matters because it defines the publishers' leverage: if unlicensed training is ultimately found unlawful, money alone may not be enough to cure it.
What USA Today Co. says OpenAI took
The plaintiffs are corporate entities behind 19 publications, including USA Today, The Tennessean, the Detroit Free Press, The Arizona Republic, the Indianapolis Star, The Columbus Dispatch, the Milwaukee Journal Sentinel and the Des Moines Register. They sued seven OpenAI entities in federal court in Manhattan on October 8.
The complaint alleges direct and vicarious copyright infringement, plus intentional removal of copyright-management information under the Digital Millennium Copyright Act. In plain terms, the publishers say OpenAI copied their journalism into training corpora without a licence, built commercial models from it, and can return outputs that reproduce or closely summarize the original work without the original attribution.
The scale figures are allegations drawn from the complaint, not an independent audit of OpenAI's current models. The publishers say their sites contributed more than 160,000 entries to WebText, the dataset associated with GPT-2, including 83,266 from USA Today. They also cite more than 122 million tokens from their domains in a 2019 snapshot of C4, a dataset associated with GPT-3 research.
Why the destruction demand matters
Publisher lawsuits against AI companies have settled into a familiar pattern: identify copies in training data, show outputs that allegedly substitute for the original article, and argue that the resulting product competes for the same readers and subscription revenue. USA Today Co.'s filing follows that playbook, but it arrives when several related cases are already concentrated in the same court and moving toward consequential rulings.
That timing gives the new complaint more weight than another isolated lawsuit. The case asks to be treated as related to consolidated copyright proceedings involving other publishers. If it joins that track, the publishers can benefit from evidence and legal arguments developed across the larger fight rather than starting from zero.
The deletion request also creates a technical problem. A training dataset can be filtered or rebuilt. A finished model does not contain a neat folder labelled USA Today that can simply be removed. Courts will eventually have to separate three different questions: whether copying into a training corpus was lawful, whether a model's outputs infringe specific works, and what remedy is technically and legally proportionate if either answer goes against the developer.
The $250 million is a negotiating position
More than $250 million is a serious claim, but it is not a bill OpenAI must pay today. The complaint seeks damages, profits or disgorgement, fees and injunctions. Each category requires legal findings and proof. OpenAI had not publicly responded when the initial reports were published.
The broader market has already shown the alternative path. Some publishers license archives to AI companies; others sue; a few do both with different companies. USA Today Co. has commercial AI relationships elsewhere, which makes this lawsuit less a rejection of AI training than a rejection of training without a deal.
The practical signal to model builders is simple: provenance debt is becoming litigation debt. A model may be technically finished while the rights to its training material remain unresolved. The practical signal to publishers is harder. A destruction demand sounds existential, but the cases will turn on evidence, fair use and remedies—not on the size of the headline.
For now, OpenAI's models are not being destroyed. A group of publishers has asked a court to make that possible. That difference is the whole story.
Sources
- [1] Digital Watch Observatory — “Newspapers owned by USA Today Co. seek over USD 250 million from OpenAI” (Oct 9, 2026)Read source
- [2] WKYU — “USA Today network, including Courier Journal, sue OpenAI” (Oct 9, 2026)Read source
- [3] Press Gazette — “News generative AI deals revealed: Who is suing, who is signing?”Read source
- [4] Runtimewire — “USA TODAY Co. sues OpenAI over training on 19 publications” (Oct 8, 2026)Read source