On October 8, 2026, one of America's largest newspaper publishers went to federal court against the company behind ChatGPT. In USA Today Co., Inc. v. OpenAI Foundation, USA Today Co. (the company formerly known as Gannett) and 13 of its affiliated newspaper companies accuse OpenAI of copying "hundreds of thousands" of their articles to train and run ChatGPT. They are asking for more than $250 million in damages. Within 48 hours the filing had been covered by Reuters, The Verge, Forbes, Bloomberg Law, MediaPost, Press Gazette, The Economic Times, and dozens of other outlets. Most headlines led with the dollar figure. The more interesting part is how the publishers built the case and what they are asking the court to order beyond money.
The case, No. 1:26-cv-08892, was filed in the U.S. District Court for the Southern District of New York. It is a copyright case (nature of suit 820) brought under 17 U.S.C. § 501, with additional claims under the Digital Millennium Copyright Act. The publishers are represented by Steven Lieberman of Rothwell, Figg, Ernst & Manbeck, the same firm that represents The New York Times in its own case against OpenAI and Microsoft.
Who is suing, and who is being sued
The plaintiffs are USA Today Co. and 13 affiliated entities that together hold copyrights in content from 19 publications. Press reports and the docket name USA TODAY, The Tennessean, the Detroit Free Press, The Arizona Republic, The Indianapolis Star, The Columbus Dispatch, the Milwaukee Journal Sentinel, The Oklahoman, the Des Moines Register, and the Louisville Courier Journal among them. For many readers in mid-sized American cities, these are simply "the paper."
On the other side are seven OpenAI entities: OpenAI Foundation, OpenAI Group PBC, OpenAI OpCo, OpenAI Global, OpenAI GP, OAI Corporation, and OAI International. Suing the whole corporate family is now standard in these cases. OpenAI's structure has changed several times, and plaintiffs want to make sure that whichever entity ends up holding the models and the revenue is bound by any judgment. Microsoft is not named as a defendant here, though it features heavily in the complaint's allegations.
OpenAI did not respond to requests for comment from Reuters, The Verge, or other outlets.
What the complaint alleges
According to reporting on the complaint by Unite.AI, Louisville Public Media, Forbes, Hoodline, and others, the publishers describe three separate kinds of wrongdoing:
- Copying to build the models. The complaint alleges OpenAI scraped the publishers' journalism "regardless of paywalls or other access restrictions" and used it as training data. It says the plaintiffs' content makes up more than 160,000 entries in WebText, the internal corpus OpenAI built to train GPT-2, including 83,266 entries from usatoday.com and 12,994 from freep.com. It also says the publications' domains account for more than 122 million tokens in C4, a filtered 2019 snapshot of the Common Crawl web archive, with about 23 million tokens from usatoday.com and 3.1 million from the Courier Journal alone. The complaint reproduces OpenAI's own published GPT-3 training mix, which weighted Common Crawl at 60 percent and WebText2 at 22 percent.
- Copying in the outputs. The publishers say ChatGPT reproduces or "repackages" their reporting in ways that substitute for reading the original. One exhibit reportedly collects output examples from GPT-5.6, and Louisville Public Media reports that one example shows ChatGPT closely mirroring a specific Courier Journal article.
- Stripping copyright management information. Under 17 U.S.C. § 1202, it is a separate violation to remove author names, titles, copyright notices, and similar identifying information from a work. The complaint alleges OpenAI used programs that removed this information from the articles it collected.
To show that OpenAI knew what it was doing, the complaint leans on the company's own words. As reported, it quotes an OpenAI research VP saying, "We train our networks to memorize the training data — that's their objective," and cites 2022 internal messages acknowledging that GPT-4 would memorize large amounts of data. It quotes ChatGPT head Nick Turley describing OpenAI's products as an "existential threat" to news publishers that "will get more and more substitutive as they get better." It also cites OpenAI's December 2023 written evidence to a UK House of Lords inquiry, in which the company said that limiting training data to public-domain works would not produce AI systems that meet current needs.
The complaint also borrows from the New York Times litigation. It cites recently unredacted documents from that case, including a Microsoft executive's description of OpenAI's use of copyrighted content as "an astonishing theft of unprecedented proportions" and possibly the "largest theft of labor in human history." It alleges that over three years Microsoft gave OpenAI a copy of the Bing search index, including the plaintiffs' content, under an initiative codenamed "Project Taxi." We covered those unsealed papers in detail when they surfaced in The New York Times Company v. Microsoft Corporation.
Why $250 million, and why "destroy the models"
The damages figure comes from two statutory menus, not from a calculation of lost subscriptions:
- Copyright statutory damages. Under 17 U.S.C. § 504(c), a copyright owner can elect $750 to $30,000 per registered work infringed, and up to $150,000 per work for willful infringement. The complaint is accompanied by an exhibit listing the publishers' copyright registrations. Those registrations are what make per-work statutory damages possible in the first place.
- DMCA damages. Under § 1203, each violation of the copyright-management-information rule carries $2,500 to $25,000 in statutory damages.
Across hundreds of thousands of articles, even a fraction of those amounts adds up quickly. The "more than $250 million" figure is best read as a floor the plaintiffs say they can justify.
The non-monetary requests may matter more. The publishers ask for a permanent injunction and the destruction of any GPT or other large language models and training sets that incorporate their content, along with a jury trial. Courts rarely order anything that drastic. Copyright law does allow courts to order the destruction of infringing copies under § 503, though, and asking for it gives the plaintiffs leverage in any settlement talks.
Where this case is headed
On the same day it filed the complaint, the plaintiffs filed a statement of relatedness asking that the case be treated as related to In re OpenAI Copyright Infringement Litigation, No. 1:25-md-03143, the consolidated proceeding in the Southern District of New York before Judge Sidney H. Stein and Magistrate Judge Ona T. Wang. That proceeding already gathers the New York Times case, The Intercept Media v. OpenAI, Ziff Davis v. OpenAI, the Authors Guild class action, and other cases. U.S. News & World Report, which sued in November 2025, was later accepted as related, and newer suits from Encyclopaedia Britannica and The Seattle Times are pending in the same courthouse.
If the court grants relatedness, USA Today will join an established pretrial process with shared discovery, rather than starting from scratch. The Times case has already forced out internal documents that every later plaintiff can quote, and this complaint quotes them heavily.
The legal question at the center of all of these cases is still fair use. AI companies argue that training transforms the material into something new. Two 2025 rulings in California, Bartz v. Anthropic and Kadrey v. Meta, found that training on books could be fair use on the records in those cases. Both judges also said that pirated sources or proof of market harm could change the result. The publishers' complaint seems written with that in mind. It stresses paywall circumvention, memorization, and outputs that "substitute" for the original, which are the facts most likely to weigh against fair use under the fourth factor, effect on the market. Not every publisher is suing, either: OpenAI signed a licensing deal in 2024 with News Corp, owner of The Wall Street Journal and the New York Post. For the publishers that do sue, part of the goal is to set the price of that kind of license.
None of these allegations has been tested in court. OpenAI has not yet responded to the complaint.
What this means for filmmakers and other creators
A newspaper chain suing an AI lab may seem distant from independent film. But the publishers' playbook applies to anyone whose work can be scraped, including scripts, finished films, trailers, stills, and scores.
Registration is what makes your rights worth enforcing. USA Today can ask for up to $150,000 per article only because it registered its works. Under § 412, statutory damages and attorney's fees are generally available only for works registered before the infringement began or within three months of publication. Register your screenplay, your finished picture, and your original score. Newspapers use the Copyright Office's group registration options to cover huge volumes of content, and a filmmaker with a body of short films or a series should look at what group options apply to them.
Your credits and metadata are legally protected. The § 1202 claim depends on copyright management information, meaning author names, titles, and copyright notices. For a film, that includes the copyright notice in the end credits, the title card, embedded file metadata, and watermarks on screeners. Keep them consistent across every file you send out. If someone strips them out to hide what they took, that can be a separate violation with its own damages.
Access controls matter. The complaint stresses that content was taken "regardless of paywalls." Password-protected screeners, expiring festival links, and DRM on delivery files are not just security measures. They are evidence that you did not consent to bulk copying, and they can weigh against a fair-use defense.
Market substitution is the argument that wins. Courts have been more willing to accept training as transformative than outputs that compete with the original. If you find an AI tool producing scenes, characters, or footage that track your work closely, document the prompt, the output, and the date. That is the same kind of evidence these publishers attached as exhibits.
Licensing is the other half of this fight. For every publisher that sues, others are negotiating deals. Rights holders with clean chain of title, documented ownership, and organized catalogs are the ones who can negotiate a license. If you can't prove what you own, you can't sell rights to it, whether the buyer is a streamer, a distributor, or an AI company.
We'll track the relatedness ruling, OpenAI's response, and any coordination with the consolidated proceeding as they happen. Follow the filings, parties, and press coverage on the case page.
This post is editorial commentary on public court filings and news coverage, not legal advice. The allegations described are claims in a newly filed complaint and have not been proven. Details come from the docket and press reports on the complaint and may be refined as the case develops.