At the federal courthouse in Manhattan, where The New York Times is suing OpenAI over the articles its reporters wrote, the government has picked a side. On Wednesday the Trump administration filed a 20-page friend-of-the-court brief arguing that training artificial-intelligence models on copyrighted works is a fair use that the courts should not narrow. The same day, Reuters reported, the United States pressed its G20 partners to allow AI systems to train on creators’ works. Two documents, one message: Washington is endorsing the data appetite of its AI industry.
The brief is blunt about the stakes as the administration sees them. Read the fair-use doctrine too narrowly, it argues, and you impede the creation and progress of American AI and damage the dynamism of the economy that depends on it. The argument echoes the industry’s own talking points: models learn from text the way people learn from books, and the copying that happens inside a training run is a means to an end, not a reproduction of the work itself.
The case has become the standard-bearer for an entire publishing industry. The Times sued OpenAI and its partner Microsoft in December 2023, accusing the companies of copying millions of articles to build the models behind ChatGPT and demanding damages along with the destruction of training data built from its archive. OpenAI answered that its use is transformative and therefore lawful. Since then, other publishers have chosen different paths, signing licensing deals with OpenAI and its rivals, while The Times held out for a court ruling.
The legal question is old and the facts are new. Fair use weighs the purpose of the copying, the nature of the work, the amount taken and the effect on the market for the original, and the closest precedent favors the technology: in 2015 an appeals court ruled that Google’s scanning of millions of books to build a search index was a fair use, a decision AI companies cite constantly. The courts that have grappled with AI training since have split. Last year one company was hit with a heavy penalty over a model built on a pirated library of books, though the ruling turned on how the data was obtained rather than on whether training itself can be fair use, and other companies have won successive rounds in similar suits.
What the government’s intervention changes is the atmosphere around those rulings. Courts are not bound by the executive branch’s views, and fair use is decided case by case, but an amicus brief from the United States carries weight on questions of public interest, and the administration’s choice to file one at all is unusual in a private copyright fight. It signals that the White House treats training data as an industrial-policy question rather than a property dispute, part of the competition with China over who builds the most capable models.
The G20 push extends the same logic into trade policy. Reuters reported that the United States used the forum to urge other governments to let AI companies train on copyrighted works, a position that puts Washington at odds with European capitals where creators hold stronger rights and AI companies face transparency obligations about the material they use. The American argument, in diplomatic language, is that heavy-handed copyright rules would cede the technology’s future to jurisdictions with looser regimes.
The stakes for the AI industry are existential in a commercial sense. A ruling for OpenAI would entrench the current generation of models, confirming that the data on which they were built was lawfully acquired and keeping the cost of the next generation where it is. A ruling for The Times would hand publishers lasting bargaining power: future licenses would carry real prices, past training runs could be revisited for damages, and the companies building the next models would have to pay for the words they learn from.
The consequences would differ even within a Times victory, lawyers caution. Copyright covers expression, not facts or ideas, and much of what models absorb from news articles is information rather than prose. An adverse decision would not erase existing systems overnight, but it would make the data behind the next generation considerably more expensive and give publishers a seat at the table where models are designed.
For OpenAI the case has always been about more than this lawsuit. The company has signaled that it will eventually sell shares to the public, and its founding bet, that the open web is fair game for learning, is now being tested in court at the same time it is being priced by investors. A government that sides with its position does not decide the case, but it tilts the ground on which the case will be decided, and it tells the industry that Washington would rather regulate the winners than unmake them.
