Does Copyright Law Lose to National Security?

Washington says access to copyrighted material may be necessary for America to win the AI race. If so, why are publishers and creators expected to finance it?

The federal government has entered the AI copyright fight, and it has chosen a side.

On Sep 1, 2026, the Department of Justice filed a statement of interest in the copyright litigation involving OpenAI, The New York Times and other authors and publishers. The government did more than offer a technical interpretation of fair use. It connected access to copyrighted training material with American economic strength, military capability and competition with China.

AI, the government argued, will play a critical role in intelligence analysis, battlefield systems, scientific discovery and economic competitiveness. Restricting the material available to American models could slow innovation and allow foreign competitors to take the lead.

Copyright litigation is no longer merely copyright litigation. It has become a national-security issue.

Read the full report (PDF)

That changes the question. If OpenAI believes its use of copyrighted material qualifies as fair use, it can make that argument like any other defendant. If the United States believes privately financed creative work is essential to a national project, then we need to discuss who should bear its cost.

Right now, the apparent answer is publishers, authors, musicians, photographers, software developers and other creators.

I am not a disinterested observer. Techstrong pays journalists, editors, analysts, producers, designers and contractors to create original reporting, research, podcasts and videos. We also pay to publish and distribute that work.

To date, Techstrong has never received a dollar from an AI company for using our content to train a model. I cannot say which companies used it, how much they used or where they obtained it. That is because model developers generally disclose so little about their training datasets that a publisher like Techstrong has no reliable way to know.

A copyright owner cannot license a use it cannot see or challenge an unlawful copy it cannot identify. The AI company controls the dataset, logs and evidence. Everyone upstream is left to guess.

The web is publicly accessible. It is not ownerless.

An Exemption Congress Never Passed

The DOJ filing does not change copyright law. It is not legislation, a regulation or a binding decision. It is nevertheless a major intervention. According to Reuters, this is the first time the federal government has entered one of the major AI copyright cases to advance a position on training.

The administration is asking the court to treat model training as extraordinarily transformative, discount broad claims of market harm and consider the danger of allowing China to advance while American developers face copyright restrictions.

Congress has never enacted a blanket exemption for AI training. It has not created a national-security exception to copyright or added competition with China as a fifth fair-use factor.

Yet the progression is difficult to ignore. The AI industry asked Washington for broad freedom to train. The White House adopted much of its strategic framing in its AI Action Plan. The administration discouraged Congress from deciding when licenses should be required. Now the Justice Department is asking the judiciary to adopt an interpretation that could produce the functional equivalent of an AI-training exemption.

The United States may decide that access to copyrighted work is necessary for American AI leadership. That would be a consequential public-policy decision. Congress should make it openly, with rules governing transparency, lawful acquisition, permitted uses and compensation.

It should not emerge indirectly from private lawsuits in which copyright owners must reconstruct secret datasets after the fact.

The Anthropic Asterisk

Anthropic agreed to pay $1.5 billion to settle claims involving books allegedly downloaded from pirate libraries and retained in a central research collection. But the settlement did not establish that every use of a book for AI training requires a license. The court had already ruled that Anthropic’s training on lawfully acquired books was fair use. The remaining exposure involved how it allegedly obtained and retained the copies.

The emerging rule sounds reasonable: Training may be fair use, but stealing the copies is not.

For a smaller publisher, that distinction may be almost impossible to enforce. How would Techstrong determine whether an AI company crawled an open page, circumvented a restriction or acquired an archive from a pirate source? How do copyright owners prove illegal acquisition when the defendant possesses virtually all the evidence?

There may not be one legal standard for large and small publishers. There may simply be one set of enforceable rights for organizations that can afford years of litigation and another for everyone else.

Train As We Say, Not As We Do

The industry’s response to Chinese model distillation makes the contradiction harder to overlook.

OpenAI and Anthropic object when Chinese companies use millions of outputs from American models to accelerate competing systems. They describe the practice as theft, free-riding and a threat to national security. Their contracts and access controls restrict customers from using those outputs to build rival models.

The legal issues are not identical. Distillation may involve fraudulent accounts, contractual violations, technical circumvention or export controls. The economic principle is harder to distinguish.

One organization invests billions of dollars creating something valuable. Another uses the resulting material without permission to reduce the cost of building a competing product. The original investor says this undermines innovation.

Authors, musicians and publishers might find that argument familiar.

When an AI company copies human-created work, it calls the process learning. When another company learns from the AI company’s outputs, it calls the process theft. When access benefits American model developers, national security supports openness. When it benefits Chinese competitors, national security demands protection.

You cannot have it both ways.

Our new Techstrong Research special report, Does Copyright Law Lose to National Security?, examines the DOJ intervention, developing case law, Anthropic’s copyright litigation, the economic impact on publishers and the compensation systems Congress could consider.

The report does not argue that AI development should be stopped or pretend that licensing billions of works would be simple. It asks a more basic question about who is being required to subsidize America’s AI ambitions.

National security can justify taking or licensing private property. It should not make that property disappear.

Download the complete Techstrong Research special report: “Does Copyright Law Lose to National Security?”