What Just Happened
The Trump administration walked directly into the biggest copyright fight in AI history this week. In a 20-page brief filed in the Southern District of New York, the U.S. government threw its weight behind OpenAI in The New York Times v. OpenAI, arguing that training large language models on copyrighted material is fair use.
The brief doesn’t mince words. “The United States has a strong interest in continuing to develop a robust and competitive artificial intelligence industry that sets the standard for the practice and procedure of AI use globally,” it reads, tying the argument directly to an executive order Trump signed in January 2025 on maintaining American AI leadership.
This isn’t a ruling. The DOJ doesn’t get to decide the case. But when the government files a statement of interest in a high-profile copyright fight between a legacy publisher and the company behind ChatGPT, it sends a signal that echoes far beyond one courtroom in Manhattan.
The Case in Plain English
Here’s the conflict in a nutshell. The New York Times sued OpenAI in December 2023, arguing that ChatGPT’s training data included millions of the newspaper’s articles, and that this amounted to copyright infringement. The Times wants “billions of dollars” in damages, according to reporting from The Verge.
OpenAI’s defense rests on a concept that has anchored copyright law for centuries: fair use. The idea is that you can use someone else’s work without permission in certain circumstances, particularly when what you’re doing is “transformative” — you’re not just copying, you’re building something new.
The Times wants the court to draw a line: training an AI on journalism isn’t transformative, it’s extraction. OpenAI’s position, echoed by the DOJ, is that an LLM reading articles to learn language patterns is closer to what a human journalist does when they read the paper every morning than it is to running a photocopier.
Why the DOJ’s Brief Matters
Statements of interest from the Justice Department aren’t binding. The judge is free to ignore them. But they carry considerable persuasive weight, and they’re part of a deliberate strategy.
The Verge’s Hayden Field reported that the Trump administration has leaned heavily on statements of interest as a policy tool, with one official calling them “incredibly” successful at advancing its aims. This brief is an extension of the same posture the administration laid out in its National AI Legislative Framework, where it argued that AI training should count as fair use.
There’s also a personal dimension. Trump is currently pursuing his own defamation suit against The New York Times, and the administration’s antipathy toward the paper is well documented. Whether that animation helps or hurts OpenAI’s case is for the judge to decide — but it’s impossible to ignore.
Not the First Time Courts Have Been Asked This Question
The fair use question in AI training isn’t new, and courts have already started weighing in. Last year, Judge William Alsup of the Northern District of California ruled on a related case — Authors Guild v. Anthropic — that produced a mixed result with an important distinction.
Judge Alsup found that Anthropic could legally train its models on lawfully purchased books. In his words, “Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different.” The comparison to human reading was direct and favorable to AI companies — though the security landscape around AI is getting its own multi-billion-dollar push, and training-data liability is part of what enterprises are paying to manage.
But the same judge also found Anthropic liable for piracy — specifically, for downloading books from shadow libraries rather than purchasing them. The $1.5 billion settlement that resulted wasn’t about training at all. It was about where the books came from. That distinction — lawful acquisition versus pirated acquisition — may end up being the real dividing line in these cases, not the training itself.
What This Means for Publishers
The Times isn’t alone. The Center for Investigative Reporting, the Chicago Tribune, and the New York Daily News have all filed AI copyright suits. Dozens of media outlets, meanwhile, have done the opposite — signing licensing deals with OpenAI, including the Associated Press, Axel Springer, and Vox Media.
The Times itself entered a licensing deal with Amazon in 2025, allowing its content to appear in Amazon’s generative AI tools. So even as it sues OpenAI, it has chosen a different path with a different company — similar to how publishers navigated the Google ad-tech breakup, choosing enforcement and partnership at the same time. That isn’t hypocrisy — it’s a signal. Publishers are testing two strategies at once: litigation to set legal boundaries, and licensing to capture value where the law is unsettled.
The DOJ’s brief, if it carries weight, tips the legal boundary in one direction. But licensing deals aren’t going away. Even if OpenAI wins the fair use argument, many publishers will still want paid partnerships — not because they have to, but because they can.
What This Means for Developers and Builders
For anyone building with AI, the most important thing to understand is that this case — and the legal questions around it — are nowhere near settled. The DOJ’s brief is a data point, not a verdict. Other AI leaders have made similar arguments about slowing down, but this is the first time the federal government has formally backed one side in court.
If the Times ultimately loses and the court affirms that LLM training is fair use, the implications are enormous. It would mean that the current model — scraping the open web, processing publicly available text, training on published works — is broadly legal. That’s the environment most AI companies have been operating in, and a favorable ruling would protect it.
If the Times wins, the landscape changes overnight. Training data becomes a liability. Publishers who are currently negotiating licenses gain an advantage. New entrants without established data pipelines face a much higher bar. The cost of building a competitive model would rise, possibly dramatically.
Neither outcome is guaranteed. The trial is in the Southern District of New York — a respected but not Supreme Court-level forum. Appeals will follow regardless of who wins. And the Anthropic precedent, while favorable on the training question, is from a different district and a different set of facts. Courts don’t move in straight lines.
The Bigger Picture: Two Competing Visions
What’s happening in this case is bigger than OpenAI and the Times. It’s a collision between two visions of how the AI economy should work.
One vision — the one the Times and many creators advocate for — treats published work as property with a price. If you want to use it to build a product, you pay for it. The law already works this way for most industries; the question is whether AI training fits inside the existing frame or needs a new one.
The other vision — the one the DOJ and OpenAI are arguing for — treats published work as part of the information environment that any learner, human or machine, is entitled to absorb. The value isn’t in the raw material but in what you build from it. Restricting access to training data, in this view, doesn’t protect creators — it protects incumbents and slows down the entire field.
These two visions aren’t easily reconciled, and the courts are going to have to pick a direction. The DOJ’s brief is an early, unusually public signal of where the current administration wants that direction to point.
Where Things Stand Now
The Times v. OpenAI case is still in its early stages. The DOJ’s statement of interest was filed on September 2, 2026. The trial hasn’t happened. The judge hasn’t ruled. Nothing is settled.
But the trajectory is worth watching. The government has staked out a position. Multiple courts have now weighed in on related questions, and OpenAIs own chief scientist has said no lab has AI fully under control. The Anthropic settlement established that piracy is actionable even when training isn’t — a practical enforcement mechanism even in a world where fair use wins. And publishers, for their part, are hedging, suing on one track while signing licenses on another.
For anyone working in AI, writing about AI, or publishing content that AI systems might train on, the next few years will define the rules of the road. This case is one of the first real attempts to write those rules into law — and the outcome will shape questions like whether we can even tell AI-generated content from human work.