On September 1, the U.S. Department of Justice sent a twenty-page document to a federal court in Manhattan supporting OpenAI in its dispute with The New York Times. According to the government, training large language models on copyrighted texts falls under the fair use exception. This is the first time the U.S. government has directly commented on the dozens of lawsuits filed by authors and publishers against companies developing artificial intelligence. A few days later, both sides submitted motions for summary judgment to the judge, meaning a ruling could be issued in the coming weeks.
Who is suing OpenAI and why
The New York Times filed its lawsuit at the end of 2023, targeting both OpenAI and Microsoft, which has invested the most money in the company. The newspaper claims that the two companies used millions of its articles to train the chatbot without permission. Initially, the newsroom asked the court to order the destruction of the models and training data containing its texts and sought billions of dollars in damages. The dispute is not only about training data. The lawsuit also alleges that ChatGPT sometimes quoted long passages from the newspaper's articles verbatim. The newsroom adds that OpenAI threatens its journalists' livelihoods and has effectively appropriated work worth billions of dollars.
OpenAI has maintained from the outset that the model does not learn texts but rather statistical relationships between words, and that such use is transformative because it creates something new. OpenAI also argued that its products do not replace the newspaper's journalism in the marketplace. The company wrote that the Times forced the chatbot to reproduce its own material using misleading prompts that violated the terms of use. The company even stated that the newspaper had hired someone to hack its products. Its defense also included the objection that some of the claims were already time-barred due to the three-year statute of limitations.
The court was not persuaded, and in 2025 the judge rejected most of OpenAI's motions to dismiss the lawsuit, allowing the majority of the newspaper's claims to proceed. Although the judge narrowed the scope of the dispute by dismissing several secondary points, the main part of the case remained on the table.
Last year, the cases were consolidated into a single major proceeding in New York. Alongside the newspaper is a group of writers, including John Grisham and George R. R. Martin. But there are many more plaintiffs. They include regional newspapers and smaller nonprofit newsrooms such as The Intercept. The New York Daily News and Chicago Tribune, publisher Ziff Davis, and the Center for Investigative Reporting have also joined.
The government's position
The Department of Justice used a format known as a statement of interest. This document is not binding, but it shows the judge how the executive branch interprets the law. In it, the government wrote that the United States has a strong interest in the court rejecting the claim that training language models on copyrighted texts infringes copyright. It based its position on both scientific progress and national security.
According to the document, the benefits and new possibilities offered by model training clearly outweigh any potential harm to competition. The government argues that the model and its outputs differ sufficiently from the original article to constitute a new work with a different purpose. ChatGPT does not seek to entertain or educate readers directly like traditional reporting, but instead generates text, translates, and completes assigned tasks. The document dismisses the objection that the chatbot can reproduce an entire article as irrelevant. According to the government, the proceedings concern the training itself, not the outputs, and ordinary use should not be assessed based on how someone might manipulate the tool in exceptional circumstances.
Another part of the argument concerns geopolitics rather than law. The government claims that a ruling in favor of publishers would threaten national security, give foreign adversaries a competitive advantage, and hinder creative and scientific progress. According to the Department, such a ruling would run counter to the fundamental principles of copyright law and undermine Trump's January 2025 executive order intended to preserve America's leadership in artificial intelligence.
Deputy Attorney General Stanley Woodward spoke about it in similar terms. On the social network X, he wrote that the president had made clear how crucial superiority in artificial intelligence is to security and economic growth. He added that this administration would not allow America to fall behind its adversaries because of a mistaken interpretation of copyright law.
The Department further claims that if OpenAI lost, all companies would have to pay licensing fees. Because of the enormous volume of published texts, this would benefit large traditional publishers the most. According to the government, creating an oligopoly in which only the largest technology companies can afford to train models is not in the public interest. The fees would therefore function primarily as subsidies for legacy media organizations. According to the document, language models instead help small newsrooms compete. Using artificial intelligence, for example, they can generate an illustrative image for an article instead of paying a photographer or an expensive licensing fee. To support its view, the government even cited The New York Times itself. It refers to the newspaper's own article stating that journalists in its newsroom use models to come up with topics and edit texts.
The plaintiffs' response
The newspaper did not hold back. Spokesperson Graham James said that the administration was siding with a handful of trillion-dollar companies against countless American creators whose work had been stolen. According to him, artificial intelligence and creators can thrive side by side. However, companies must pay fairly for content, as copyright law requires. An effort to allow businesses to take content without permission and without compensation would, according to the spokesperson, undermine the sustainability of human creativity. A healthy society depends on it, and the models themselves need it to function.
Attorney Matt Topic, who represents The Intercept in the proceedings, describes it as an unprecedented and uncompensated transfer of rights from newsrooms to technology companies.
Courts remain divided
The government's support is not binding on the judge, and the U.S. judiciary is divided on the issue. Last year, two judges in San Francisco reached different decisions. William Alsup described Anthropic's use of books as highly transformative. Although Vince Chhabria ruled in Meta's favor two days later, he also warned that in many situations training could not be considered fair use. Chhabria specifically cautioned that generative models could flood the market with content that competes with human creators.
Anthropic ultimately reached a settlement with authors for approximately one and a half billion dollars, which the court approved. For the company, it was only a partial victory. Although the judge recognized that training on books was sufficiently transformative to qualify as fair use, other handling of pirated materials was not. In May, another lawsuit was filed in New York in which publishers Hachette, Cengage, and Elsevier accuse Meta of using copyrighted content.
How the lawsuits are proceeding
OpenAI, Microsoft, The New York Times, and the group of authors filed opposing briefs on Friday. Each side is asking Judge Sidney Stein to rule in its favor without a trial. OpenAI relied specifically on Alsup's decision and described training as one of the most transformative processes humanity has ever experienced. The company claims that it extracts broad, uncopyrightable statistical patterns of language from texts, not the copyrighted expression itself. Microsoft added that after years of discovery, the situation is clear and that neither training nor the use of models in products replaces books.
The authors disagree. In their filing, they wrote that artificial intelligence is flooding the book market across the entire industry and that fair use cannot shield such a threat to creative authorship. News publishers make a similar argument, pointing out that ChatGPT draws readers away from their websites.
The ruling will almost certainly reach an appeals court, but it will establish the first benchmark for dozens of other disputes awaiting a decision. The New York Times can afford such a delay because it has already spent more than thirty million dollars on artificial intelligence litigation. Only a few small and nonprofit newsrooms are suing OpenAI because legal proceedings are too expensive for them, so they are leaving the battle to the best-funded newsroom in the country.
Sources: wsj.com, reuters.com, nytimes.com and theguardian.com
