US government sides with OpenAI on issue of training LLMs on copyrighted material

The Trump administration has stepped into the New York Times v. OpenAI fight with a 20-page brief that takes OpenAI’s side on a key question: can AI companies train models on copyrighted material without asking first? The government says yes, or at least that copyright law should not be read in a way that blocks that kind of training. Its argument is tied to a broader national-interest claim: the U.S. wants a strong AI industry and wants to keep “global leadership in artificial intelligence.”

That matters because the models behind ChatGPT, Claude, Gemini, and similar systems are trained on huge collections of books, articles, and other published work. Publishers have pushed back hard, saying AI firms have no right to feed their material into these systems without permission. The legal fight has largely turned on fair use, the copyright doctrine that can allow unlicensed use when a court sees it as transformative enough.

The administration’s brief says limiting LLM development through a bad reading of fair use would choke off creative and scientific progress, and it links that directly to American prosperity and economic mobility. It’s a political statement as much as a legal one. And it lands at a time when the courts have been, on balance, more friendly to AI companies than many rights holders expected.

A year ago, Judge William Alsup ordered Anthropic to pay $1.5 billion in a copyright settlement to writers, but the punishment was for using illegal shadow libraries, not for training its models on the books themselves. Alsup even compared the process to a reader learning from a book and then making something new. That analogy is now doing a lot of heavy lifting across Silicon Valley and Washington alike.