Paywalled Content AI Lawsuits Surge as USA Today Seeks $250M

The paywall has become the main fault line

For three years, the main legal question in generative AI was abstract: is training a model on copyrighted text fair use? This autumn the fight has narrowed to something more specific and more damaging for defendants. Publishers are now asking how AI companies got the material in the first place. They point to subscription walls, crawler blocks and copyright notices that they say were bypassed or removed. In this round of litigation, the paywall is the central issue, and the evidence assembled so far favors the publishers.

The latest filing came on October 8, when USA Today Co. and affiliated newspapers sued OpenAI in Manhattan federal court. The complaint covers work from 19 publications and seeks more than $250 million.16 The complaint says OpenAI's WebText dataset held more than 160,000 entries from the plaintiffs' websites, 83,266 of them from usatoday.com. It also says the publishers' domains made up more than 122 million tokens in the C4 dataset.5 The publishers argue that they protected their journalism with paywalls, terms of service, robots.txt files and copyright notices, and that OpenAI did not filter paywalled material out of its training sets.5 They want statutory damages, an injunction and an order under the Copyright Act to destroy GPT models and training sets that contain their content.5

The suit is part of a wave. Emmerich Newspapers and other publishers sued Microsoft and OpenAI in Mississippi federal court on October 2.11 The Seattle Times and Newsday filed in the Southern District of New York in early September, alleging that paywalled articles were scraped into the datasets behind ChatGPT, Copilot and Bing.3 More than 30 local newspaper owners, with nearly 400 titles between them, have also sued the two companies.2 A tracker that follows the issue counts 144 AI copyright cases.11

What the unsealed Times filings changed

The biggest reason for the shift is the material Judge Sidney Stein unsealed in September in The New York Times' case against OpenAI and Microsoft. He is weighing summary judgment motions in that case.7 Reporting on the documents agrees on the main points. Microsoft applied-science director Brent Hecht wrote in an internal document that AI scraping was "the largest theft of labor in human history."4 When OpenAI researcher Nick Ryder told Greg Brockman about a "hack to get around nytimes paywall," Brockman answered "ah nice."1 OpenAI's ChatGPT chief Nick Turley wrote that AI products were "largely substitutive, period."10

The filings also put numbers on the issue. Internal Microsoft data reportedly showed that Copilot's answer-style search could cut click-throughs to the Times by as much as 93 percent compared with standard Bing.6 The Times says a dataset derived from Common Crawl held more than two million documents from nytimes.com. It also says a dataset built through Microsoft's "Project Mango" contained at least 160,903 unique works from the publisher plaintiffs.6 Researchers allegedly removed copyright notices from training data so the models would not repeat them in their answers.6

The most consequential statement may be Satya Nadella's deposition. He testified that anything paywalled "should be licensed by anyone who wants to use it" for training or grounding. He added that if he had known OpenAI trained on paywalled material, he would have used Microsoft's right to make OpenAI retrain its models.1 A Microsoft spokesperson later said Nadella was speaking about broad principles, and the company said Hecht's remarks reflected his own view, not company policy.48 Those clarifications do little to change the record. Plaintiffs now have a defendant's CEO stating the licensing rule they are asking the court to enforce.

The coverage does differ in tone. Some outlets treated the documents as admissions of guilt; one carried the headline "Elite Crime Spree."7 TechCrunch and Firstpost were more careful. They noted that much of the new information comes from the Times' own brief, and that the underlying exhibits are still sealed and cannot be checked independently.16 That caution is warranted. These are allegations framed by advocates, not findings of a court.10 Even so, the more careful outlets reported the same quotes, and neither company has disputed that the words were written.

Why circumvention matters more than training

The legal reasoning explains why publishers keep returning to paywalls. In Bartz v. Anthropic, the court found that training on copyrighted books was fair use, but that keeping pirated copies was not. That split led to a $1.5 billion settlement, about $3,000 per work.12 The case showed that the way material was acquired can create liability even when training itself is protected. A paywall works as the newspaper industry's version of that line. If the content was taken from behind one, the defendant's conduct looks less like reading the open web and more like knowingly taking something it had not paid for.

Recent rulings point the same way. On October 9, Judge Seeborg denied Apple's motion to dismiss a DMCA anti-circumvention claim under Section 1201 in Ted Entertainment v. Apple.11 A judge also found it plausible that Perplexity conspired with third-party scrapers to get around Reddit's access controls, and refused to dismiss that case.2 News Corp's suit against Brave says the browser maker disguised its crawlers to slip past publisher blocks.2 Taken together, these cases suggest that claims about technical barriers can get past early motions even where the fair use question is still open.

The removal of copyright management information (CMI) is a less certain theory. USA Today says OpenAI's extraction tools, called Dragnet, Newspaper and Gutentag, stripped out copyright notices and sometimes author names.5 Meta won on a similar DMCA claim in Kadrey.12 The Times plaintiffs, though, now have internal evidence that the stripping may have been intentional, and that could make the theory work this time.

The first appellate ruling, and how far it reaches

On September 29, the Third Circuit affirmed the judgment in Thomson Reuters v. Ross Intelligence. It is reported to be the first federal appellate decision on fair use in AI training.1718 Ross used material derived from Westlaw headnotes to train a competing legal research tool. The courts held that the headnotes are copyrightable and that the use was not fair.17 The district court rested on market harm: Ross was building a direct substitute for the product it copied.13

The case also matters for legal tech. Thomson Reuters had refused to license Westlaw to Ross because the two would compete.13 Companies that build AI research tools on top of proprietary legal databases now face a binding appellate precedent that rewards licensing and penalizes building around a refusal. Ross has said it will ask the Supreme Court to take the case.17

The reaction to the ruling has been uneven. Some commentary treated it as a general verdict against AI training. Others pointed to footnote 7, where the court said Ross's system "cannot generate original expression," so the concerns in the generative AI cases "do not apply here."19 One tracker said that within days some outlets had overstated the footnote as a protected zone for generative AI, when the court had only reserved that question.18 Both readings go too far. The ruling does not decide the generative cases. But the court recognized harm to a "rapidly developing" market for licensing headnotes as AI training data.19 That market-harm theory carries over to newspapers. The Times documents, with OpenAI's own executive calling its products "substitutive," are the kind of evidence a court applying Ross's market analysis would focus on.

Policy pressure from Washington

The federal government is backing the AI companies. The Justice Department filed a Statement of Interest in the Times case arguing that a ruling for the newspaper would hurt local newsrooms, which it says can use AI to compete, and could threaten national security by slowing AI development.3 The local newspapers suing OpenAI and Microsoft are a direct counterpoint to the argument that AI helps small publishers.3 The administration's position does not change the Copyright Act, but it shows that the policy fight will continue whatever courts decide on summary judgment.

Licensing deals and lawsuits

Most of the industry is now either negotiating or suing. Brazil's Folha settled with OpenAI by signing a deal, and Microsoft signed its first Asia-Pacific news agreement with Australia's Nine.2 Axel Springer's 2023 agreement already let ChatGPT summarize otherwise paywalled content.2 USA Today's complaint uses deals like these as evidence. It argues that OpenAI's paid licenses show the company knows journalism has commercial value.5

The most likely outcome is that paywalled content ends up being paid for, through settlements, licenses or court judgments, even if training on the open web stays legally defensible. Of the 248 AI cases one tracker follows, only 25 have been decided and 18 have settled.17 Defendants have been settling to avoid precedent. With Nadella's testimony and Brockman's "ah nice" in the record, the Times case may be the one where settling stops being optional.