USA Today Co. Sues OpenAI Over AI Training, Seeking $250 Million — and the Destruction of Its Models

USA Today Co., the company formerly known as Gannett, sued OpenAI on Thursday for copyright infringement over the use of its newspapers' content to train artificial-intelligence models, seeking more than $250 million in damages and — in a remedy rarely requested and never granted at this scale — a court order requiring the destruction of every GPT and other large language model incorporating its journalism.

The 79-page complaint, filed in the U.S. District Court for the Southern District of New York as case 1:26-cv-08892, names 14 corporate plaintiffs owned by USA Today Co. covering 19 publications, including USA Today, the Indianapolis Star, The Tennessean, The Columbus Dispatch and The Oklahoman. It names seven OpenAI entities as defendants — OpenAI Foundation, OpenAI GP LLC, OAI International Inc., OpenAI OpCo LLC, OpenAI Global LLC, OAI Corporation and OpenAI Group PBC. Microsoft, OpenAI's largest investor, is conspicuously not a defendant. Reuters reported the filing the same day, identifying the case number 1:26-cv-08892.

The complaint brings three counts: direct copyright infringement under Section 501 of the Copyright Act, vicarious infringement, and intentional removal of copyright-management information under Section 1202 of the Digital Millennium Copyright Act. The plaintiffs allege OpenAI copied hundreds of thousands of their articles without permission, ingested them repeatedly to train its models, reproduced their journalism in model outputs, and stripped the copyright notices from the copies it made.

The complaint's thesis is blunt: "OpenAI never asked permission. It never offered payment. It simply took the USA TODAY Plaintiffs' work and used it to build products worth hundreds of billions of dollars." The plaintiffs, represented by Steven Lieberman of Rothwell Figg, allege a traffic-diversion motive: OpenAI's head of ChatGPT acknowledged that once ChatGPT answers a question, there is "no good reason to click" on a link to the underlying source, one passage says.

The dataset math: 83,266 articles and 122 million tokens

Unlike many AI-training complaints, this one pleads its dataset case in concrete numbers. The filing alleges the plaintiffs' content comprises "more than 160,000 entries in the WebText dataset, including 83,266 entries from usatoday.com." It separately alleges the publications' domains account for "over 122 million tokens" in C4, a filtered English-language subset of a 2019 Common Crawl snapshot, with usatoday.com alone contributing roughly 23 million tokens.

The complaint names a current-generation example: it says a GPT-5.6 model, asked to find and summarize the Indianapolis Star's July 17, 2026 article "FBI raids home, business of Westfield developer," produced "an extensive multi-section summary paraphrasing and following the same structural organization as the original article." The filing says its examples are "a few illustrative and non-exhaustive" ones, with full versions attached as Exhibit B.

The maximalist remedy: money, an injunction, and model destruction

The damages demand is "in excess of $250 million." The complaint notes the statute allows up to $150,000 per willful infringement plus up to $25,000 for each violation involving stripped copyright-management information — arithmetic that makes the nine-figure demand straightforward to reach across hundreds of thousands of alleged copies.

But the damages number may not be the most consequential demand. The plaintiffs ask the court, under Section 503(b) of the Copyright Act, to order the destruction of "all GPT or other LLM models and training sets that incorporate the USA TODAY Plaintiffs' content." Copyright scholars have long treated model destruction as the nuclear option in AI litigation — a remedy the statute provides but that no court has imposed on a frontier model. The prayer for relief also seeks a permanent injunction and a jury trial.

The conspicuous absence: Microsoft is not a defendant

The most notable litigation choice may be whom the complaint leaves out. Microsoft — OpenAI's largest investor and a co-defendant alongside OpenAI in the New York Times and Authors Guild cases — is not named, despite the complaint's reliance on a Microsoft executive's description of OpenAI's conduct as "an astonishing theft of unprecedented proportions" and perhaps "the largest theft of labor in human history." TSN Media, which reviewed the court record, confirmed that Microsoft is not a defendant. The complaint does allege OpenAI exchanged copies of plaintiffs' content with Microsoft, including Bing-index material and training datasets, but pleads no claim against it.

A tag-along timed to the consolidated case

The plaintiffs filed a statement of relatedness asking that the case be treated as related to In re OpenAI, the multidistrict copyright litigation before Judge Sidney H. Stein in the same court — the consolidated case that includes the New York Times, the Authors Guild, the Daily News newspapers, the Center for Investigative Reporting, the Intercept and Ziff Davis. But relatedness is a request, not an outcome: as of a docket review on the morning of Oct 9, no order accepting relatedness had issued, according to clearon-ai's RECAP-based review.

The timing is strategic regardless. Both sides in the consolidated case filed summary-judgment motions on Sept 4, 2026, per the Authors Guild's account of its own motion, with a ruling not expected until 2027, according to press accounts — and the USA Today complaint draws on the consolidated case's own discovery record, including the executive quotation and the traffic-diversion admissions. The Digital Watch Observatory, tracking the filing, counted 14 corporate plaintiffs covering 19 outlets and a 79-page complaint seeking over $250 million. The Justice Department weighed in on the consolidated case in September with a brief supporting OpenAI and Microsoft, according to multiple press accounts, arguing the litigation implicates scientific advancement, economic growth and national security.

What to watch: silence, relatedness, and the 2027 ruling

Three things matter next. First, OpenAI's answer: as of Oct 9, OpenAI had not publicly responded to the lawsuit, Crypto Briefing reported — and no response had surfaced by Sunday. A docket review by FourWeekMBA found the case file contained only the plaintiffs' filings and court notices. Second, the relatedness decision, which determines whether this case rides the consolidated track toward the 2027 summary-judgment ruling or proceeds on its own docket. Third, whether other publishers follow: the Kentucky outlets' own reporting notes the Louisville Courier-Journal is a named plaintiff whose content contributed 3.1 million tokens to C4, per WKYU — a reminder that the per-domain math in this complaint is a template any peer of the 19 papers could copy.

The $250 million is the headline; the model-destruction ask is the story — because if a court ever granted it, every frontier lab's training pipeline would have to be rethought.