The DOJ reiterated its training stance yesterday, in a filing submitted as part of an ongoing infringement showdown between the New York Times and OpenAI as well as Microsoft. In motion since late 2023, that case, having received a third amended complaint last month, is straightforward enough.

The way the plaintiff sees things, the defendants built generative models “by copying and using millions of The Times’s copyrighted news articles” before capitalizing on the text to pump out “substitutive products without permission or payment.”

And in the opposite corner, Microsoft and OpenAI are adamant that the described training process constitutes fair use; “Microsoft further admits that its position has been and continues to be that its conduct is protected by the fair use doctrine,” the appropriate company wrote when answering the suit two days ago.

As some will recognize, this position isn’t new for the Trump administration, which in March 2026 published a “National Policy Framework for Artificial Intelligence.” Therein, the White House emphasized the belief “that training of AI models on copyrighted material does not violate copyright laws” – while urging Congress to let the courts “resolve this issue.”

Back to the filing, then, the Justice Department noted its assessment that “[t]he copying of protected text articles as part of training an LLM is a use of a different kind or character that is ‘transformative—spectacularly so.’”

According to the legal document, “[t]he purpose of the copying (to build an intelligent, interactive model) differs in kind from the purpose of the copied work (to use language to directly entertain or educate a reading audience).”

“But LLM training is its own use. It does not involve any substitution. … In fact, training does not reveal anything to the public at all—it simply creates a copy of a protected work in order to teach an LLM to recognize relationships between data and adapt to new information,” the DOJ wrote.

“The potential for future outputs that might cause market harm is simply not relevant to evaluating an LLM training use under the required use-by-use analysis.”

Running with this point, though the brief’s chief arguments probably won’t prove popular in an industry where many training-related copyright suits are underway, the government struck a different tone with regard to outputs.

(Nor are the arguments resonating with the New York Times, which in a statement accused the administration of “siding with a handful of trillion-dollar AI companies at the expense of the countless American creators whose work they stole.” On the other hand, Re:Create executive director Brandon Butler applauded the government for having “rightly intervened to ensure that fair use can continue to power American leadership on AI innovation.”)

Important generally and because recent days’ Suno copyright suits have cited multiple allegedly infringing outputs, the distinction is worth keeping in mind moving forward.

So is the broader threat presented by generative models developed in nations with less robust (i.e., nonexistent) IP protections – a threat that the government touched on in its brief.

“Rules of law that make it significantly more difficult to develop a robust AI industry in the United States therefore threaten national security and give a competitive advantage to foreign adversaries who are not so encumbered,” the DOJ wrote.

At the same time, if U.S.-based AI music giants are sued into oblivion – even when they’re willing to change their detrimental policies and ink rightsholder-lucrative licensing deals – will we be left with smaller-scale licensed platforms and a collection of massive international competitors that flout IP laws with impunity?

Perhaps. With billions upon billions of dollars in anticipated revenue on the horizon in China, the industry powers that be remain conspicuously silent about years-old AI music generators like Melodio as well as new platforms such as Alibaba’s Happy Shrimp Music.

Put one way, we know comparatively little about these platforms’ training processes, and particularly when it comes to Melodio, we don’t lack pressing output-related questions. Furthermore, each appears quite easy to access from the U.S. if so inclined.