The U.S. government has entered one of the most consequential legal battles over artificial intelligence, arguing that training AI models on copyrighted material can qualify as fair use.
In a filing submitted on September 1, the Justice Department backed OpenAI in its copyright dispute with The New York Times, saying restrictions based on an overly broad reading of copyright law could hurt U.S. technological progress, economic development and national security.
The filing does not settle the dispute. That decision remains with the federal court. But the government’s intervention could add significant weight to OpenAI’s argument as courts grapple with how decades-old copyright rules should apply to generative AI.
DOJ Puts AI Innovation First
The Justice Department’s position is built around the economic and strategic importance of artificial intelligence. Reuters reported that the government argued an incorrect interpretation of copyright law could hinder America’s ability to develop AI and compete with other countries.
The administration also connected AI development to national security. In its court filing, the DOJ argued that AI systems can generate public and creative benefits and that those benefits can overlap with national-security interests.
That makes the case bigger than a dispute between one technology company and one publisher. The question before the court could affect how AI developers obtain the enormous quantities of text and other material needed to train increasingly capable models.
Still, the government’s filing is an argument to the court, not a change in copyright law. Whether particular uses of copyrighted material qualify as fair use remains a case-specific legal question.
Why The NYT Lawsuit Matters
The New York Times sued OpenAI and Microsoft in December 2023, accusing the companies of using millions of its copyrighted articles without permission to develop AI products. The Times has argued that the technology companies are benefiting from journalism in products that can compete with publishers’ own businesses.
The litigation has already produced important procedural rulings. On April 4, 2025, Judge Sidney H. Stein rejected several attempts by OpenAI and Microsoft to have claims dismissed, including claims involving direct and contributory copyright infringement. At the same time, the judge dismissed certain other claims, including common-law unfair competition claims.
Crucially, that ruling did not establish that training AI models on copyrighted material is inherently unlawful. The court was deciding motions to dismiss, with factual allegations generally treated as true for purposes of that stage of the case.
Licensing Offers Another Route
The legal uncertainty has also encouraged a parallel commercial approach: licensing.
OpenAI has entered agreements with news organisations rather than relying exclusively on litigation to determine how publishers’ content can be used. In January 2025, OpenAI said it had partnered with nearly 20 media organisations, representing more than 160 news outlets and hundreds of content brands.
A later OpenAI announcement concerning The Washington Post said the company had partnerships with more than 20 news publishers, reaching more than 160 outlets.
These deals demonstrate that licensing can coexist with the broader fair-use debate. But they do not establish that licensing is legally mandatory for all AI training. That remains one of the central questions being contested in court.
Copyright Law Faces AI Test
The U.S. Copyright Office has been studying these questions since launching its AI initiative in 2023. Its notice of inquiry attracted more than 10,000 comments by December 2023, reflecting the breadth of concern among technology companies, creators, publishers and other stakeholders.
The Office released a pre-publication version of its Part 3 report on May 9, 2025, specifically examining generative AI training. The report does not create a blanket rule permitting or prohibiting AI training. Instead, it examines how existing copyright principles and possible policy approaches apply to the practice. The Office continues to identify Part 3 as a pre-publication version, with a final version still forthcoming.
That leaves the latest DOJ filing as an important intervention, but not a final answer.
For AI companies, a broad fair-use interpretation could preserve access to large training datasets without requiring negotiations with every copyright owner. For publishers and creators, the opposite outcome could strengthen the case for permission-based licensing and compensation.
The New York Times case therefore sits at the intersection of copyright, AI economics and U.S. technology policy. The eventual ruling could help define whether copyrighted content used in AI training is primarily treated as protected material requiring permission, or as material that can, in certain circumstances, be transformed through machine learning under fair-use principles.
Also read: New York City Bans Student AI Use Through 8th Grade, Allows Limited High School Access
