A Justice Department filing in the New York Times case gives AI companies fresh ammunition on training, even as music labels press their own copyright fights.
The Trump administration has taken a side in the AI copyright wars, and it is not the one music companies wanted to see.
In a filing Tuesday in the New York Times lawsuit against OpenAI, the Department of Justice said AI companies do not violate copyright law when they train models on written works without a license. The filing is advice to the court, not a ruling, but it is the first time Washington has stepped into one of the copyright cases now piling up against AI developers.
That matters well beyond the newspaper case. The same fair use fight sits at the center of the music industry’s lawsuits against Anthropic, Suno and Udio. The DOJ’s position is limited to text, not songs, but the logic could still shape how judges think about training, outputs and market harm.
Washington backs training, not licensing
The 20-page “Statement of Interest Of The United States” was signed by Associate Attorney General Stanley Woodward. It says the government has “a strong interest in this Court rejecting any argument that training LLMs on copyrighted texts violates copyright law.”
The filing leans on Donald Trump’s AI policy, citing executive orders from January 2025 and June 2026, along with the administration’s National Policy Framework for Artificial Intelligence, published in March. That framework says the “training of AI models on copyrighted material,” by itself, “does not violate copyright laws.”
The DOJ focuses on the two questions that usually decide fair use disputes: how transformative the new use is, and whether it harms the market for the original. On the first point, the government argues that copying text to train a model like ChatGPT is “a use of a different kind or character,” and “extraordinarily transformative.” On the second, it says training copies do not substitute for the original because training “does not reveal anything to the public at all.”
The filing also warns that forcing large AI companies to pay licensing fees to publishers could tilt the market toward legacy media outlets simply because they have so much content. The government says it is not in the public interest for the biggest tech companies to face “an oligopoly on LLM training due to licensing entry barriers that function primarily as large subsidies for old mainstream media companies.”
Why music lawyers are watching anyway
The DOJ is careful to say its brief is about “copyrighted texts,” “written works,” and “text articles.” A footnote limits the reasoning to the OpenAI case and related suits brought by “book authors and publishers.” Recordings and compositions are not mentioned.
Still, fair use is fair use. Music companies have built their AI cases around the same basic question: whether training on copyrighted material without permission is lawful, and whether the outputs prove the training was harmful in the first place. A judge in the Suno case could read the government’s view of the test and decide it has broader value.
The DOJ splits AI development into three stages: acquiring the material, training the model on it, and generating outputs. It says each stage raises separate copyright questions. The government is defending only the middle one.
That distinction matters because the music industry is attacking all three. The first stage — how the material was obtained — has already become a weak spot for AI companies. In the book authors’ case against Anthropic, Judge William Alsup ruled in 2025 that downloading books from pirate libraries was not fair use, calling it “straightforward piracy but at massive scale.” Anthropic later settled with those authors for $1.5 billion in September 2025.
Two of the four counts in Sony Music Publishing and Warner Chappell Music’s new suit against Anthropic also concern torrenting. The DOJ’s filing says nothing about that.
The market harm argument gets hit
The sharper clash is over market harm. In the 2025 Meta case brought by book authors, Judge Vince Chhabria suggested that AI outputs could flood the market with competing works, and that developers may therefore need to pay copyright holders for the right to use their materials, even at the training stage. Music rightsholders have leaned on that theory ever since. Lawyers call it market dilution.
In a brief filed March 30, the RIAA, NMPA, A2IM, SoundExchange and four other groups asked a court to reject Anthropic’s fair use defense in a case brought by UMG, Concord and ABKCO, making a similar market harm argument.
The DOJ now calls Chhabria’s reasoning “deeply flawed.” It says he “improperly collapsed LLM training and LLM outputs into a single continuous use.” Training and outputs are separate legal questions, the government argues, and what a model produces has no bearing on whether training it was lawful.
That is the line music lawyers will have to wrestle with. If a court accepts it, labels and publishers may find it harder to argue that a flood of AI-generated tracks proves the training itself was infringing. The government’s filing does not decide that question. It does, however, give AI companies a new federal argument to carry into the next round.
For music, the fight just got more complicated.
