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

Amanda Silberling reports the Trump administration filed a 20-page brief in the New York Times' suit defending OpenAI's unlicensed use of copyrighted material to train LLMs. The brief says the United States has a strong interest in a robust AI industry that sets global standards, and that it is critical to 'retain global leadership in artificial intelligence,' citing an executive order President Trump signed last year.

The legal fight is over fair use and whether training is 'transformative' enough. The brief warns that constraining LLM development under a misunderstanding of fair use would thwart creative and scientific progress and hinder American prosperity. Silberling notes cases have largely favored AI companies. Judge William Alsup ordered Anthropic to pay $1.5 billion because it pirated books from shadow libraries, not because training was illegal; he wrote that the LLMs trained on works 'not to race ahead and replicate or supplant them — but to turn a hard corner and create something different,' like a reader aspiring to be a writer. The new brief is not a ruling. The case sits in the Southern District of New York. The brief's authors do not have jurisdiction. The intervention can still carry weight.

Silberling's contribution is to put the executive order in the citation and Alsup's 'hard corner' sentence next to it. The administration is trying to nationalize a metaphor judges have already used: training is reading. A 20-page brief that quotes the President's own EO is not subtle. It is asking Stein to treat OpenAI's corpus as a matter of state.

No jurisdiction is the legal fact. Weight is the political one. SDNY does not work for the White House. It also does not ignore DOJ paper on a file this large. The Anthropic split remains the controlling story on the ground: training survived, the acquisition pipeline did not. A brief about Times articles is about the Times' theory of harm, not about whether OpenAI torrented a library.

Signal Desk's read: this is industrial policy entering a copyright docket because Congress has not written an AI-training statute. Fair use is being asked to carry 'American prosperity.' That is a lot of freight for a doctrine built around criticism and parody. If Stein buys it, every publisher suit gets harder. If he does not, the administration still got the quote it wanted — global leadership — into the clip.

ChatGPT, Claude, and Gemini all trained on 'incomprehensibly massive' copyrighted stores, as Silberling writes. A Times-only theory was always going to become a sector theory. The brief just said that out loud.

Publishers argue unlicensed training is infringement. Labs argue it is fair use because the output is a new statistical machine, not a substitute article. Courts have been warmer to the labs on training and colder on how the bytes were obtained. The administration has now picked a side in the Times case without being a party.