What the government filed
On September 1, the United States entered the consolidated OpenAI copyright litigation through a 20-page Statement of Interest. The Justice Department argues that using copies of written works to train a language model is different in purpose from publishing those works for readers and can therefore be highly transformative. It asks the court to reject a categorical argument that training on copyrighted text is infringement. A Statement of Interest presents the government's legal view; Judge Sidney Stein remains free to accept, narrow or reject it.
Three stages, three questions
The filing separates acquisition, training and output. Material may first be collected and stored, then copied through the training process, and later influence generated responses. The government defends the training stage addressed in its brief, not every way data might have been obtained or every output a user might elicit. It acknowledges that reconstructing and distributing an original work can raise a different question. That division matters because a favorable view of training does not automatically excuse pirate-source allegations or outputs that substitute for protected articles.
Why music cases do not follow automatically
The filing speaks about text, written works and related claims by authors and publishers. It does not analyze sound recordings or musical compositions. Music Business Worldwide notes that judges in cases involving Anthropic, Suno and Udio may still encounter the government's reasoning about transformation and market harm, but they must apply it to different facts and asserted rights. The brief also challenges the idea that a hypothetical licensing market for training should control the analysis. Whether any court adopts that position—and how it weighs creator compensation against competition—remains unresolved.