The DOJ Just Rewrote the Rules on AI Training Data

The Government Just Weighted Its Side of One of Tech’s Biggest Legal Battles

The Trump administration filed a letter in federal court on Tuesday effectively telling a judge to side with OpenAI in the New York Times’ long-running copyright case. The Department of Justice did not merely express sympathy for the AI industry — it framed the outcome as a matter of national competitiveness.

The government argued there is a “strong interest” in whether training AI systems on copyrighted material counts as fair use, because the question directly affects whether the United States can maintain “global leadership in artificial intelligence.” That is not neutral language. It is the government taking a side in a lawsuit that could determine how the AI economy is built.

Why This Matters Beyond the Courtroom

This is not a narrow dispute about newspaper archives. The DOJ’s position, if adopted by the court, would establish a precedent that echoes across every AI company in the world. Writers, photographers, musicians, and software developers have all seen their work used to train large language models without permission or payment. A ruling that training is fair use would shield that practice for the foreseeable future.

The government made two specific arguments that go to the heart of fair use doctrine. First, it called AI training “extraordinarily transformative” — meaning the systems do not merely copy input but fundamentally repurpose it. Second, it argued that OpenAI’s models do not significantly compete with the New York Times’ articles, another factor courts weigh in fair use cases.

Both claims are contested. Publishers and authors say AI outputs increasingly replicate the tone, structure, and substance of protected works. They argue the competitive harm is real even if a model does not publish news stories verbatim.

The Comparison That Reveals the Stakes

Perhaps the most provocative argument in the government’s letter is its comparison of AI training to human creative practice. The DOJ noted that conflating what AI training does with what AI tools produce would create “problematic implications for copyright law generally.” It pointed to the fact that teenage writers — including a young Joan Didion, who reportedly copied out Hemingway stories by hand — learn by studying the work of others.

If that analogy holds, then training an LLM on billions of words is no different in kind from a human apprentice reading widely. If it does not hold, then the comparison is an evasion.

The difference, of course, is scale and speed. Didion spent years mastering a craft through deliberate imitation. OpenAI’s systems ingested vast swaths of published work in days. Whether that distinction matters legally is exactly what the courts will have to decide — and the DOJ wants them to decide in favor of the AI companies.

Who Wins, Who Loses

OpenAI and Microsoft win immediately. They face fewer legal headwinds, lower compliance costs, and a clearer path to continue training without licensing deals. Their competitors — smaller AI startups that lack the resources to negotiate with every publisher — also benefit from a broad fair-use ruling.

The New York Times and other publishers lose. They spent years building the legal case that their content has value and that unauthorized use violates copyright. A government brief arguing against them undercuts years of effort.

Authors and creators lose too. The Authors Guild, which filed its own lawsuit against OpenAI, called the DOJ letter “replete with faulty arguments and a gross misunderstanding of the fair use doctrine.” Mary Rasenberger, the guild’s chief executive, said the nonprofit is “extremely disappointed.”

There is a third group that loses quietly: the public. If AI training is declared fair use without compensation, the incentive structure for creating original work shifts. Content creators may receive nothing for the raw material that powers trillion-dollar companies.

The Political Angle Is Hard to Ignore

This is a Trump administration move. The president has positioned himself as pro-AI and anti-regulation, and the DOJ’s letter reads like a policy document dressed in legal language. The government tied copyright doctrine to economic competitiveness and American prosperity — themes that align with the administration’s broader industrial strategy.

That alignment matters. Judges may be more inclined to defer to the executive branch when it frames a legal question as an economic one. The DOJ did not ask the court to rewrite copyright law; it asked the court to interpret it in a way that supports industry growth. Whether that interpretation is legally sound is secondary to the political signal being sent.

What Happens Next

The judge in the case, Sidney H. Stein, is not bound by the government’s letter. As IP lawyer Evan Brown noted, Stein can and should decide based on the law. But Brown also acknowledged that courts “take it quite seriously” when the Department of Justice weighs in. That carries weight.

The broader implications extend far beyond this single case. There are dozens of ongoing copyright lawsuits involving AI companies. A favorable ruling here would strengthen the defense in every other one. Meta already won a technical victory in Kadrey v. Meta last year, though the judge left the door open for future claims. The DOJ’s intervention gives those AI companies even more cover.

Publishers and creators are likely to push back hard. They will argue that the government is prioritizing corporate interests over constitutional copyright protections. They may seek to appeal whatever ruling comes out of this case. And they could push Congress to act — though legislative reform on this issue faces an uphill battle given the administration’s stance.

The Real Question Is About Power

At its core, this case is about who gets to profit from the digital economy. The old model — authors publish, publishers distribute, readers pay — is being upended by AI systems that can generate competent text without paying anyone. The DOJ’s letter makes clear which side of that transformation the government favors.

The government’s argument that constraining AI development would “thwart creative and scientific progress while hindering American prosperity” is compelling on its face. But it assumes that the current trajectory — unlimited training data at zero cost — is the only path to progress. That is a policy choice, not a legal necessity.

Whether that choice will hold in court remains to be seen. But the fact that the administration felt compelled to intervene so forcefully tells you something important: the AI companies needed this, and they knew it.