USA Today Sues OpenAI for $250M Over AI Training on 19 Papers

A New Plaintiff With a Familiar Grievance

USA Today Co. has sued OpenAI. It is one more name on a long list of news organizations that say the ChatGPT maker built its business on their journalism without paying for it. The company and 13 affiliated entities filed the complaint on Thursday, October 8, in federal court in New York. They allege that OpenAI copied hundreds of thousands of articles from 19 publications to train and run its models, and that ChatGPT then reproduced or repackaged that reporting for its users.1 The case is captioned USA Today Co v. OpenAI Foundation, No. 1:26-cv-08892, in the Southern District of New York. Steven Lieberman of Rothwell Figg Ernst & Manbeck represents the publisher.3

The papers involved stretch well beyond the flagship national title. They include The Tennessean, IndyStar, The Columbus Dispatch, The Oklahoman, the Detroit Free Press, The Arizona Republic, the Milwaukee Journal Sentinel and The Courier-Journal, among others.15 The complaint puts the stakes in blunt terms, asserting that OpenAI's commercial success depends on large-scale copyright infringement and that the company took the publisher's content to build products worth hundreds of billions of dollars.2

OpenAI did not respond right away to requests for comment.3 The allegations are only claims at this point. The $250 million figure is what the plaintiffs are asking for, not anything a court has awarded.4

What the Complaint Actually Alleges

Coverage of the basic facts is consistent. Every major account reports a damages demand above $250 million, a filing in Manhattan federal court, and an allegation that hundreds of thousands of articles were copied.238 The details in the complaint show more about the plaintiffs' strategy than the headline number does.

According to the filing, the papers account for more than 160,000 entries in WebText, the dataset OpenAI assembled to train GPT-2. It also says they make up more than 122 million tokens in C4, a 2019 Common Crawl snapshot, with 23 million of those tokens coming from usatoday.com.1 Those figures matter because proving what went into a model has usually been the hardest part of these cases. Developers rarely disclose full training sets, so plaintiffs tend to build their case from circumstantial evidence and then fight over discovery.10 Tying the claims to named, partly public datasets gives USA Today a stronger starting point than earlier plaintiffs had.

The damages theory has two parts. The plaintiffs seek up to $150,000 per willfully infringed work, which is the statutory maximum. They also seek up to $25,000 for each instance in which OpenAI allegedly stripped copyright management information.1 The complaint argues that OpenAI used tools designed to remove those copyright markers.1 The remedies go further than money. The publisher asks the court to order the destruction of GPT models and training sets that contain its content, and it demands a jury trial.16

The most pointed material is OpenAI's own internal language. The complaint quotes Nick Turley, OpenAI's head of ChatGPT, saying publishers face an "existential threat" from the company's products and that those products are "largely substitutive." It also cites an engineer's observation that users won't click on links no matter how prominently they appear, and internal documents describing ChatGPT as a "modern newsstand."1 The plaintiffs further argue that OpenAI's paid licensing deals with other news organizations show the company knew a license was required.1

Why the Market-Harm Argument Is the Real Target

In my view, this complaint is aimed at the fourth fair-use factor: the effect of the use on the market for the original work. Analysts following the New York Times case have argued that this factor has become the decisive issue in AI training disputes. Under that reading, a publisher does not need to prove that ChatGPT destroyed its subscriptions. It needs to show that a licensing market exists, that OpenAI takes part in that market, and that unlicensed training undercuts it.13 OpenAI's existing deals with News Corp, Axel Springer and the Associated Press arguably serve as evidence that the market exists.13

USA Today's complaint appears built around that theory. The "substitutive" quotes support the market-substitution argument. The references to licensing deals support the existence of a licensing market. The traffic and subscription harms support the claim of lost revenue.1 The case law points the same way. In Thomson Reuters v. Ross Intelligence, a court rejected fair use for a legal-research AI that competed directly with Westlaw, the source of its training material. That ruling matters to legal-tech vendors that build on proprietary databases.13 In Kadrey v. Meta, Meta won, but on a thin record, and the court suggested that a market-dilution theory might be the stronger path for plaintiffs who can prove harm.13 Bloomberg Law has reported that the reach of the first AI copyright appeal depends on the market-impact question.18

The Anthropic litigation adds a separate risk: how the training data was obtained. In Bartz v. Anthropic, the court treated training as transformative but viewed the acquisition of pirated copies separately and unfavorably.13 Anthropic later agreed to a $1.5 billion settlement covering roughly 482,000 books.14 Coverage differs on the timing of that settlement. Politico places it in July, while The New York Times reported it in May.1419 Either way, the settlement showed that AI defendants can face very large payouts even where training itself might be protected. USA Today's claims about stripped copyright information seem designed to create a similar liability that does not depend on how the fair-use question is resolved.

The Bigger Fight in the Same Courthouse

The new suit arrives at a sensitive point in the main case. The Times' 2023 lawsuit against OpenAI and Microsoft has been consolidated with complaints from other publishers into a multidistrict litigation before Judge Sidney Stein in the Southern District of New York.1819 Earlier rulings in that proceeding went largely against OpenAI. In October 2025, the court refused to dismiss output-based claims where plaintiffs alleged outputs a jury could find substantially similar to their works. In January 2026, it ordered OpenAI to produce 20 million output logs.12 The parties have since briefed fair use at summary judgment.1317 The news plaintiffs have also sought sanctions over alleged discovery misconduct.18

In September, the federal government became involved. The Justice Department filed a statement of interest supporting OpenAI's fair-use defense, calling AI leadership a national-security interest and arguing that LLM training is transformative because it learns statistical patterns rather than copying expression.1419 According to The New York Times, it was the first time the department had taken a position on AI training and copyright.19 The DOJ also argued that mandatory licensing fees would favor the largest tech companies and "disproportionately benefit legacy media outlets."18 The publishers responded that the administration may be biased in OpenAI's favor and that the department had brushed aside the Copyright Office's contrary conclusion on fair use.11

The two sides do not even agree on the economics. Ziff Davis CEO Vivek Shah estimated that annual royalties across the news industry would cost roughly $20 billion. He compared that with the $750 billion OpenAI expects to spend on computing infrastructure by 2030 and called licensing a "rounding error."17 OpenAI has argued that its scraping fell under an implied license because publishers did not originally block AI crawlers through robots.txt.17

The overlap in legal representation is worth noting. Rothwell Figg is among the firms representing the news plaintiffs in the consolidated case and is now leading USA Today's suit.318 That suggests a coordinated publisher strategy and not a series of unrelated grievances.

Where the Coverage Agrees and Where It Doesn't

On the facts of the filing, reporting is uniform. The differences are in emphasis. Reuters-based accounts present the case as one of dozens of training-data suits against OpenAI, Anthropic and Meta.37 Forbes focused on the internal OpenAI quotes and the demand to destroy models.1 Industry commentary stressed caution: nothing has been decided, and both licensing and litigation remain open as strategies.4 One explainer said plainly that its authors had seen only press accounts and not the complaint itself.10

Some commentary has suggested this case could finally settle whether training on news is fair use.6 That expectation is probably misplaced. Fair use is decided case by case, and older suits are further along.10 Judge Stein's eventual ruling in the consolidated litigation will almost certainly shape this case more than anything in its own early stages. One analysis predicts a split outcome: training treated as transformative in general, with liability attaching to specific outputs, retained copies or acquisition conduct.13 That analysis also expects any ruling to be appealed to the Second Circuit, which would push finality toward 2028.13

The Policy Stakes

Courts are not the only place this is being decided. The Supreme Court's March 2026 denial of certiorari in Thaler v. Perlmutter confirmed that human authorship is still required for copyright protection.12 That settles the output-ownership question while the training question remains open. The administration's AI policy framework says creators should be protected from infringing outputs without undercutting lawful innovation. The president has also encouraged Congress to consider licensing frameworks or collective rights systems.18 Separately, EU transparency rules require general-purpose model providers to publish summaries of their training content. Those disclosures could feed future U.S. litigation.13

My read is this. USA Today is unlikely to be the plaintiff that breaks new legal ground. What it adds is weight: more registered works, more local newspapers that can describe concrete community harm, and more of OpenAI's internal statements on the record.110 That increases the cost of OpenAI losing the larger fight while the company is relying on the federal government's support to win it. For a publisher with an archive this large, a lawsuit costs little compared with what it could gain at the negotiating table. The likely result is more publisher suits, and more licensing deals reached while those suits are pending.