The US government tells the New York Times court that training a model on copyrighted text is fair use
The Trump administration has filed a 20-page statement of interest in The New York Times' copyright suit against OpenAI and Microsoft, arguing that training a large language model on copyrighted text is fair use. The Justice Department separates the copying done during training from what a finished model puts out, and says a narrower doctrine would hold back research and the American AI industry. The fair-use question is the court's to answer.
Artificial Intelligence··Night
The executive branch enters the fair-use fight on the defendant's side
The US government filed a 20-page statement of interest in the US District Court for the Southern District of New York, where The New York Times says OpenAI trained ChatGPT on its journalism without permission. In the filing US attorneys write that the newspaper wants fair-use doctrine narrowed to exclude the training of OpenAI's large language models, that such a result would be inconsistent with basic copyright principles and would hamper the progress of science and useful arts, and that the country has a strong interest in developing a competitive AI industry. The newspaper has separately accused OpenAI of withholding material during discovery. The filing offers an argument, not a ruling; the fair-use question is still the court's to answer.[1], [2]
The department separates the copying in training from the model's output
The Justice Department made the same argument for the consolidated case, which gathers the newspaper together with other rights holders. In its account, entire works are copied during training but never made publicly available, and the outputs often if not always lack substantial similarity to the originals, so a blanket theory of market harm that treats the two as one use is legally wrong. The filing reaches for Joan Didion, who as a teenager copied out Hemingway's stories to learn how his sentences worked: under the logic of the Kadrey ruling, the department argues, Didion could have faced liability every time she published. The text adds that people create original works with the help of large language models, that those models are already helping researchers across fields achieve major breakthroughs, and that liability for training alone would stifle the creativity copyright law exists to protect. US attorneys also write that the fair-use inquiry hinges on the specific facts and uses at issue in each case. The US Copyright Office reached the opposite conclusion in a report rejecting blanket fair use for AI training, noting that such systems work from perfect copies and generate at a speed and scale beyond human creation.[3], [2]
The case has run since December 2023, and the administration has used such filings before
The New York Times sued OpenAI and Microsoft in a Manhattan federal court in December 2023, saying millions of its articles were used without permission to train models such as GPT-4, citing billions of dollars in damages and asking for the destruction of models trained on that work. The administration has leaned on statements of interest in private litigation before and set out the same fair-use position in its National AI Legislative Framework, while Trump pursues a defamation suit against the same newspaper. In a 2025 decision a judge found that Anthropic's training on lawfully purchased books was legal while the company remained liable for piracy, and Anthropic settled with authors for 1.5 billion dollars. Dozens of outlets, among them The Associated Press, Axel Springer and Vox Media, have signed licensing deals with OpenAI, and the newspaper licensed its own content to Amazon in 2025.[2], [3]
