USA Today publisher sues OpenAI for copyright infringement

USA Today Co., the publisher formerly known as Gannett, and a group of its newspapers sued OpenAI for copyright infringement on 8 October. Crypto Briefing and Let’s Data Science reported the case was filed in the US District Court for the Southern District of New York. The complaint itself was not available to this record, so the account below rests on press reports of its contents, and the allegations are unproven.

According to those reports, the publisher alleges that OpenAI used “hundreds of thousands” of articles from its titles, including paywalled ones, to train ChatGPT without permission. The complaint reportedly points to OpenAI’s WebText dataset, saying it contains more than 160,000 entries from the plaintiffs’ domains, of which about 83,300 came from usatoday.com. Outlets differ on how many newspapers are plaintiffs: Crypto Briefing counts 18 affiliated papers and Let’s Data Science says content from 19 publications is at issue. Let’s Data Science also reports that the defendants include the OpenAI Foundation and related OpenAI entities.

The suit also goes beyond training. MLex reported that USA Today accuses OpenAI of scraping its sites to train models and of repackaging the content in outputs to users, and said OpenAI “deliberately targeted news content”. The Desk reported that the complaint cites remarks attributed to Nick Turley, described there as the OpenAI computer scientist in charge of ChatGPT, that AI-generated summaries could reduce traffic to other websites, along with an OpenAI engineer who allegedly wrote that users were unlikely to follow source links however prominent they were. The plaintiffs argue ChatGPT’s summaries cut web traffic and advertising revenue. Crypto Briefing and Let’s Data Science report claims of direct and vicarious infringement and of removing copyright management information; the relief sought is more than $250 million in damages, an injunction, and, according to Crypto Briefing, destruction of models trained on the publisher’s work.

OpenAI had not commented in the coverage reviewed, and the court has not ruled on any claim.

The suit joins the New York Times case against OpenAI, in which the US Justice Department filed in support of AI training as fair use in September, and other newspaper actions. It makes the same core claim, that scraping journalism to train commercial models requires a licence, and will face the same unresolved question of how courts treat training under fair use.