Why the DOJ Backed OpenAI and What It Means for AI's Money Problem

A brief that isn’t a ruling, but carries weight

The Trump administration filed a 20-page brief in the New York Times v. OpenAI copyright case. It sides with the chatbot maker. It does not decide the case. Judge Lewis Kaplan in the Southern District of New York does that. But the government’s intervention is a signal — one that cuts across every AI company using scraped content, and one that publishers are already reading as both a threat and an opportunity.

The brief’s thesis is blunt. “Constraining LLM development under a misunderstanding of fair use doctrine would thwart such creative and scientific progress while hindering American prosperity and economic mobility,” it says. The argument wraps AI training in the language of national competitiveness, invoking a Trump executive order on AI leadership. The message to courts, to Congress, and to every media company watching: the federal government sees unlimited access to copyrighted works as essential to winning the AI race.

That framing matters more than most readers will realize.

Why this brief changes the economics, not just the law

Copyright law has always had a fair-use doctrine — the legal principle that lets you use protected material without permission in certain circumstances: criticism, commentary, teaching, parody. AI companies have argued that training large language models on vast corpora of published works is transformative, much like a reader absorbing ideas from books and turning them into something new. Judge William Alsup of the Northern District of California said exactly that last year when he dismissed copyright claims against Anthropic over its training data. “Like any reader aspiring to be 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,” he wrote.

But the New York Times case lives in New York, not San Francisco. Judge Kaplan has not ruled on the merits of the fair-use argument yet. The government’s brief tips its hand: it wants the next judge to see the same path. If Kaplan agrees, OpenAI’s training methodology gets a green light that echoes through every startup building on public data. If he does not, the Trump administration’s position becomes a political reference point — cited in briefs, quoted in hearings, weaponized in lobbying.

Either way, the brief shifts the terrain. The question is no longer purely whether fair use applies. It is now also about whether the United States wants to make fair use apply, as a matter of industrial policy.

The publishers’ leverage was always there — they just needed a reason to use it

Here is what the brief obscures: even if OpenAI wins this case, the economics of AI training are already changing. Publishers are not waiting for a court verdict to renegotiate the terms. They are negotiating now, and they are winning deals that reshape how models get built.

The New York Times has licensed its content to OpenAI, Microsoft, and other companies for seven-figure sums. Shutterstock has done the same, building a licensing business around training data. The key insight is simple but consequential: if publishers can extract revenue from AI companies today regardless of the fair-use outcome, then the legal question becomes less existential and more strategic. AI companies still want to train on unlicensed material — it is cheaper and more expansive — but they are already paying for access anyway. That creates a new market dynamic where licenses become the default, and fair use becomes a fallback rather than a right.

The Trump brief treats this as a problem. It frames licensing as a barrier to American AI leadership. But the reverse is also true: licensing is a market signal that copyrighted content has value, and once that signal is established, it is harder to unwind.

Who wins, who loses, and what happens next

If Kaplan rules for OpenAI, the immediate effect is legal cover for every AI company still operating on scraped data. The long-term effect is slower movement toward comprehensive licensing frameworks, because the economic incentive to pay publishers weakens when you can claim fair use. If Kaplan rules for the Times, the government’s brief becomes a record of opposition but not a blocker. The more likely path is a messy middle: no sweeping precedent, but enough ambiguity to keep both sides investing in lobbying and settlement negotiations.

The Anthropic settlement is worth watching as a counter-example. The company agreed to pay $1.5 billion to a group of writers, but the fine was for using illegal shadow libraries to access the books, not for the training itself. Judge Alsup drew a distinction between piracy and fair use that left the door open for licensed training to become the industry standard. The Trump brief does not address that distinction directly, which means the licensing pathway remains intact regardless of how this case resolves.

The real stakes are structural. The government is betting that unchecked access to training data will keep the United States ahead of European and Chinese competitors. But Europe is moving toward regulation that treats training data licensing as mandatory, and China has its own state-directed model of data collection that does not rely on copyright exemptions at all. The American approach — whatever Kaplan decides — will either accelerate a licensing market or entrench a scraped-data economy, and both outcomes carry global implications.

Publishers who have held back from licensing their work out of principle now face a simpler calculus: every month without a deal is a month of free training on their content. The New York Times has already proved you can have both a lawsuit and a licensing revenue stream. That model is spreading. The question is whether other publishers will follow it — or wait for a court to give them something for free.