The Trump administration has entered the copyright fight between The New York Times and OpenAI, filing a 20-page brief that supports the AI company’s position on using copyrighted works to train large language models.
The government’s core argument is economic and strategic: restricting LLM development based on what it characterizes as a mistaken application of fair-use doctrine could impede scientific progress, prosperity and U.S. leadership in artificial intelligence.
The filing does not decide the case. The lawsuit remains before the U.S. District Court for the Southern District of New York. But it gives OpenAI a notable federal-policy ally in litigation that could help define the operating rules for generative AI.
What the government is arguing
LLMs behind products such as ChatGPT, Claude and Gemini are trained on extremely large collections of text and other material, including books, articles and media protected by copyright. Publishers and creators have argued that using that material without a license is infringement.
The key legal question is whether training is protected by fair use. That doctrine allows unlicensed use in certain circumstances, with courts assessing factors including whether the use is sufficiently transformative.
The administration’s brief frames model training as an activity that should not be constrained in ways that undermine the country’s competitive position. It cites President Donald Trump’s executive order on removing barriers to American leadership in AI and says the U.S. has an interest in building a robust domestic AI industry that can set global standards.
Why it matters for operators and builders
The case is not just about OpenAI’s historical training practices. It concerns the cost structure and product roadmap of the broader AI sector.
If courts broadly accept a fair-use defense for training, frontier-model developers may be able to continue building models on large corpora without needing comprehensive licenses for every source. That would preserve an approach that favors companies with substantial compute, data-engineering capacity and legal resources.
A ruling that sharply limits unlicensed training, by contrast, could raise data-acquisition costs, elevate the importance of licensed datasets and create barriers for smaller model developers. It could also increase the value of publishers’ archives and accelerate commercial licensing markets.
For enterprises building on third-party models, the outcome may eventually affect more than vendor pricing. It could shape indemnification terms, provenance commitments, data-use restrictions and the availability of specialized models trained on licensed industry material.
The line courts may still draw
Recent AI copyright cases have offered some encouragement to model makers, but they do not establish that every method of acquiring or using data is protected.
In a separate matter involving Anthropic, Judge William Alsup approved a $1.5 billion settlement tied to books used from illegal shadow libraries. The issue was piracy in obtaining the material, rather than the legality of model training itself. Alsup characterized the training use as different from simply replicating or supplanting the underlying works.
That distinction is operationally important. Even if training can qualify as fair use, companies may still face serious exposure over how copyrighted material was sourced, stored and documented.
What to watch next
The immediate question is how much influence the government’s brief has on the New York federal court. It is an intervention, not binding law, and the fair-use analysis remains fact-specific.
Builders should watch for judicial treatment of two separate issues: whether training is transformative, and whether model outputs can substitute for protected works. Rights holders, meanwhile, will focus on evidence that models reproduce content or weaken markets for original reporting, books and other media.
For now, the brief signals a clearer federal policy preference for keeping U.S. AI development moving. It does not remove the legal uncertainty—or the need for companies to maintain disciplined data sourcing and governance.
