Key Takeaways

• The Justice Department argues that training AI models on copyrighted works can qualify as fair use and should not automatically be treated as copyright infringement.
• The government warns that restricting AI training could sharply increase development costs, reduce competition and slow American technological innovation.
• The New York Times maintains that OpenAI and Microsoft used its journalism without permission to create products that compete with the newspaper.

By Samuel López | USA Herald

The U.S. Department of Justice has stepped directly into one of the most consequential copyright battles of the artificial intelligence era, backing OpenAI’s position that using copyrighted material to train large language models can constitute fair use.

In a Statement of Interest filed September 1 in federal court in Manhattan, the DOJ argued that training AI models on copyrighted material, standing alone, does not violate copyright law. The government warned that adopting a rule effectively prohibiting such training could “severely hamper” scientific and technological progress.

The filing comes in sprawling copyright litigation that began when The New York Times sued OpenAI and Microsoft in 2023, accusing the technology companies of using millions of copyrighted Times articles without authorization to help develop generative AI systems.

The government’s filing does not decide the case, nor is the court required to follow it. But it marks the federal government’s first major intervention into the growing wave of copyright lawsuits challenging the use of protected works to train generative AI.

The Legal Fight Over Fair Use

At the center of the dispute is Section 107 of the Copyright Act, which directs courts to evaluate four factors when determining whether an unauthorized use of copyrighted material is nevertheless lawful as fair use.

Those factors examine the purpose and character of the use, the nature of the copyrighted work, how much of the work was used and the effect of that use on the market for the original. Fair use is therefore not a blanket exemption for AI companies. It remains a fact-specific legal analysis.

The DOJ nevertheless takes a strong position on AI training itself.

According to the government, copying material during the training process serves a fundamentally different purpose from publishing or selling the original articles. The filing calls large-language-model training “extraordinarily transformative,” reasoning that models learn relationships and patterns from enormous datasets in order to respond to future prompts rather than simply functioning as repositories of the underlying works.

That distinction could prove critical.

The DOJ also separates training from what an AI system may later output. If a model reproduces substantial protected portions of an article, book or other copyrighted work, that output could raise a different infringement question. The government argues those situations should be evaluated independently rather than used to characterize the entire training process as unlawful.

The New York Times Pushes Back

The Times sees the issue very differently.

The newspaper argues that OpenAI and Microsoft used valuable journalism without paying for it and then developed commercial products capable of competing for the same audience. A Times spokesperson said the government was effectively siding with enormously valuable AI companies at the expense of American creators, arguing that AI developers should compensate publishers for copyrighted material used to build their products.

The dispute therefore reaches beyond whether copying technically occurs during AI training. It raises a much larger question — when does technological transformation outweigh the economic rights copyright law gives creators?

Why Authors Guild v. Google Matters

The DOJ relies heavily on the Second Circuit’s 2015 decision in Authors Guild v. Google, which upheld Google’s creation of a searchable database generated by scanning millions of copyrighted books.

That court concluded that copying can be transformative when the material is used for a substantially different purpose and does not become a meaningful substitute for the original work.

The Justice Department argues AI training follows similar principles. According to its filing, the relevant market question is not simply whether new technology creates economic competition, but whether the challenged copying produces a substitute for protected expression.

That issue could become one of the defining copyright questions of the decade.

A ruling requiring developers to obtain licenses for virtually every copyrighted work contained in massive training datasets could dramatically increase both licensing costs and the administrative burden of building AI systems. Conversely, a broad fair-use ruling could substantially limit the ability of publishers, authors and other creators to demand compensation for works used during training.

For courts, the challenge will be drawing a line between protecting the economic incentive to create original work and allowing new technologies to build upon existing knowledge.

Whatever the court ultimately decides, this litigation is no longer simply a fight between a newspaper and an AI company. It is becoming a test of how American copyright law will operate in an era when machines can learn from enormous portions of the world's published knowledge.

About the Author

Samuel López is an independent reporter for USA Herald, where he has covered legal, technology and investigative matters since 2022. A longtime legal and insurance professional with more than two decades of industry experience.

Stay With the Stories Shaping What Comes Next

If you value independent reporting that goes beyond the headline, join readers who follow USA Herald for investigative coverage, major legal developments, emerging technology and national issues that deserve a closer look. Subscribe to the USA Herald newsletter and get our latest reporting delivered directly to you.

👉 Follow USA Herald on X at RealUSAHerald