Justice Department Backs OpenAI Fair-Use Defense in Copyright Case
A 20-page filing argues that training large language models on copyrighted text is highly transformative, while leaving questions about AI outputs for the court.
The Justice Department is urging a federal court to treat the training of large language models on copyrighted text as fair use, a position that could shape litigation brought by news organizations against OpenAI and Microsoft.
The department filed a 20-page statement of interest on Sept. 1, 2026, under 28 U.S.C. § 517. The consolidated cases are before U.S. District Judge Sidney H. Stein in the Southern District of New York.
The filing says the United States has a strong interest in rejecting the argument that training large language models on copyrighted texts violates copyright law. It describes the training process as “transformative—spectacularly so,” because models use written works to identify statistical relationships rather than reproduce the original material’s expressive content.
The Justice Department separates AI development into three stages: collecting copyrighted works, training a model and generating outputs. Its analysis focuses on copying during training, while recognizing that an output that reconstructs and distributes protected material could raise different legal issues.
The dispute turns on whether copying works into training datasets is protected by fair use. Courts generally examine the purpose of the use, the nature of the copyrighted work, the amount copied and the effect on the market for the original.
News publishers argue that AI-generated answers and summaries can compete with journalism and weaken current or potential licensing markets. Organizations including The New York Times are suing OpenAI and Microsoft over alleged copyright infringement tied to the use of news content in AI development.
The Justice Department also argues that requiring licenses for training data could raise barriers to entry and favor large established media companies. It says those requirements could place U.S. AI companies at a disadvantage against foreign competitors.
The filing does not resolve the cases or establish that every AI-training practice is lawful. The court must still examine the evidence, including whether the copyrighted works were lawfully acquired and whether particular outputs infringed copyright.
