The U.S. Justice Department has stepped into one of the world’s most closely watched artificial-intelligence copyright cases, backing the argument that training large language models on written works can qualify as fair use.
The statement of interest, filed on September 1 in federal court in New York, does not decide the case. It gives the court the federal government’s view as OpenAI, Microsoft, The New York Times and other rights holders fight over whether AI companies may use copyrighted articles and books to train generative AI systems without permission or payment.
For publishers, software companies, creators and regulators, the filing turns a private lawsuit into a broader signal about U.S. AI policy. The Justice Department told the court that the United States has a strong interest in a competitive AI industry and argued that treating model training itself as copyright infringement would risk slowing innovation, competition and national-security uses of AI.
The Times sharply rejected that position. Associated Press reported that the newspaper said the administration was siding with large AI companies at the expense of creators whose work helped make those systems possible.
What The DOJ Filing Says
The government’s argument focuses on the training stage of large language models, not every possible output those systems may produce.
In the filing, the Justice Department says training involves copying written works so a model can learn statistical relationships among words, facts, syntax and concepts. It argues that this use is different in character from the original purpose of a newspaper article or book, which is to inform, analyze or entertain readers directly.
That distinction matters because fair use is assessed by looking at the purpose of the use, the nature of the copyrighted work, the amount used and the effect on the market for the original. The DOJ argues that courts should analyze training separately from outputs that might reproduce protected expression.
The department also warns against treating broad market competition as the same thing as copyright substitution. Its position is that an AI system generating a new answer in the same general category as journalism is not automatically a substitute for the copied article unless it reproduces protected expression in a way that competes with the original.
The filing does not say all AI behavior is lawful. It acknowledges that specific outputs could raise separate legal questions if they reconstruct and distribute protected material. That leaves room for disputes over guardrails, memorization, search-style summaries and examples in which a system may return material too close to a source work.
Why Publishers Are Fighting Back
The New York Times sued OpenAI and Microsoft in December 2023, arguing that their systems used Times journalism without authorization and could compete with the newspaper’s reporting. The case was later folded into wider OpenAI copyright litigation involving publishers and authors.
The publisher argument is not only about copying during training. News organizations say generative AI products can weaken the business model that funds original reporting by answering user questions directly, reducing traffic to source sites and making it harder to sell subscriptions, advertising or content licenses.
That issue has become more urgent as AI summaries move into search, browsers, productivity tools and phones. If readers receive answers inside an AI interface, publishers may lose the click, the subscriber relationship and the advertising impression even when their reporting helped train or inform the system.
Some media companies have responded by signing licensing deals with AI developers. Others have sued. The legal question is whether copyright law requires permission for the training use itself, for particular outputs, or for neither unless the system reproduces protected material closely enough to substitute for the original.
Why The Case Matters Globally
The OpenAI copyright fair use fight is being watched well beyond the United States because American courts are likely to shape the business rules for much of the AI market.
If OpenAI and Microsoft win on training, AI companies would gain a stronger legal foundation for building models from large bodies of publicly available text, while publishers would be pushed toward voluntary licensing, product partnerships, technical controls and claims focused on specific infringing outputs.
If the Times and other rights holders win, the economics of model development could change quickly. Large AI firms may need broader licensing budgets, smaller developers could face higher entry barriers, and publishers may gain more leverage over premium archives, real-time news feeds and protected databases.
The policy tension is similar to issues in recent GDU coverage of the FSB’s warning on frontier AI and financial stability and the Anthropic Pentagon AI ruling. Governments are trying to encourage AI adoption while setting rules for competition, safety, public-sector use and accountability.
There is also a search-market angle. GDU previously covered OpenAI’s teen-safety launch, which showed how AI companies are moving from general chatbots toward more controlled product experiences. The copyright case asks whether the information supply chain behind those products remains sustainable if AI answers replace visits to original sources.
What Comes Next
The DOJ filing is advisory, not binding. Judge Sidney H. Stein must still weigh the parties’ evidence and legal arguments, including the Times’ allegations about copying, outputs, market harm and unfair competition.
The timing matters because the litigation is moving toward major rulings on fair use and liability. Courts have already begun drawing distinctions between lawful training, unlawful acquisition of pirated works, and outputs that may reproduce protected expression. The New York case could become one of the most important tests for news content specifically.
For AI developers, the immediate effect is political as much as legal. The federal government is signaling that it sees AI training as part of a national innovation and security strategy, not merely a private commercial practice.
For publishers and creators, the filing raises the stakes. If courts adopt the DOJ’s reasoning, rights holders may need to focus on proving substitutive outputs, negotiating access to premium or real-time content, and building products that preserve direct reader relationships.
The case will not settle the global AI copyright debate by itself. Europe, India, Japan and other major markets are still developing their own rules around training data, transparency, text-and-data mining, licensing and creator compensation. But the New York litigation now carries a clear policy message from Washington: the U.S. government wants courts to protect room for AI training while leaving narrower infringement claims for specific harmful uses.
