US Government Backs AI Training as Fair Use

The US Department of Justice has told a court that AI companies do not violate copyright law by using written works to develop their models without a license. The DOJ filed its 20-page 'Statement of Interest' on Tuesday, September 1, in the copyright lawsuit brought by The New York Times against OpenAI. Signed by Associate Attorney General Stanley Woodward, the document advises the court to reject arguments that this development process for large language models is unlawful. The filing represents the first time the US government has intervened in the growing number of copyright cases against AI developers.

The Justice Department's position rests on two executive orders from President Donald Trump, issued in January 2025 and June 2026, and his National Policy Framework for Artificial Intelligence from March. That framework states that using copyrighted material for model development, in itself, does not violate copyright laws. The DOJ brief tackles the two key questions in fair use analysis. First, it argues that copying text for model development is extraordinarily transformative and represents a use of a different kind or character. Second, it contends that a copy used in this process does not act as a market substitute because the process does not reveal anything to the public at all. The government adds that requiring AI firms to pay licensing fees to publishers like the Times would disproportionately benefit legacy media outlets due to the sheer volume of their written publications. It argues against creating an oligopoly on LLM development through licensing barriers that would function as large subsidies for old mainstream media companies.

To be clear, the DOJ's filing specifically addresses copyrighted texts, written works, and text articles. A footnote limits its reasoning to cases brought by book authors and publishers. Recordings and musical compositions are not mentioned across its 20 pages. However, the core legal principle of fair use applies broadly. Judges considering the defenses of music AI companies like Suno and Udio may review the government's interpretation. The DOJ splits AI model creation into three distinct stages, defending only the middle phase where models learn from ingested content. The music industry's lawsuits, including those filed by major labels and publishers against Anthropic, Suno, and Udio, attack all three stages. On the first stage, acquisition, the AI industry has already faced setbacks. In 2025, Judge William Alsup ruled in a book authors' case that Anthropic's downloading of books from pirate libraries was straightforward piracy but at massive scale and not fair use. Anthropic settled that case for $1.5 billion in September 2025.

The DOJ's filing directly challenges a legal theory central to the music industry's argument. In 2025, book authors lost a fair use case against Meta over AI development. However, the judge in that case, Vince Chhabria, suggested that AI outputs carrying the potential to flood the market with competing works might mean developers should generally need to pay copyright holders for the right to use their materials. This concept, called market dilution, forms the basis of music rightsholders' claims. In a March 30 brief, groups including the RIAA, NMPA, and A2IM asked a court to reject Anthropic's fair use defense on similar market harm grounds. The DOJ now calls Chhabria's reasoning deeply flawed, arguing he improperly collapsed LLM development and LLM outputs into a single continuous use. The government contends development and outputs are separate legal questions, and what a model produces has no bearing on whether its creation was lawful. For the third stage, outputs, the DOJ offers no defense. It concedes that certain uses may not be transformative if the LLM reconstructs and disseminates an original copyrighted work. This is precisely the ground Universal Music Group and Sony Music have staked out against Suno and Udio, arguing AI-generated songs directly compete with the recordings used in model creation. Music publishers make a parallel argument about Anthropic's Claude reproducing lyrics on demand. The New York Times criticized the administration's position. On Wednesday, September 2, a spokesperson for the paper, Graham James, said the Administration's proposal to let companies take that content without permission or compensation would undermine the sustainability of the human-created content that a healthy society depends on.