Media
THE NEW YORK TIMES WANTS TO PROTECT ITS ARTICLES. WASHINGTON SAYS AI MAY NEED THEM.
America’s biggest AI copyright fight has acquired a new participant: the United States government.
There is something wonderfully clarifying about watching a philosophical argument become an industrial policy.
For years, the dispute over artificial intelligence and copyright sounded almost academic.
Can an AI company train a model on copyrighted books, articles, images and music without paying every rights holder?
Is training “copying”?
Is it transformative?
Is it fair use?
Should creators be compensated?
Should AI companies need permission?
Lawyers argued. Artists protested. Tech companies insisted the future depended on access to data.
Now Washington has entered the room.
And Washington has picked a side.
The Trump administration has filed a brief supporting OpenAI in its copyright battle with The New York Times and other newspapers, arguing that training large language models on copyrighted texts generally qualifies as fair use.
The government’s reasoning is revealing.
This is no longer merely about whether OpenAI owes a newspaper money.
It is about American technological power.
Scientific progress.
Economic growth.
And national security.
That changes the argument considerably.
The Newspaper Says: These Are Our Articles
The New York Times sued OpenAI and Microsoft in 2023, accusing them of using millions of Times articles without permission to build products that can sometimes reproduce, summarize or compete with the newspaper’s own journalism.
The basic complaint is easy to understand.
Journalists research stories.
Editors edit them.
Photographers photograph them.
Publishers pay salaries, legal bills, foreign bureaus, servers and all the other expenses required to produce journalism.
Then an AI company collects enormous quantities of that work and uses it to build a commercial product worth billions.
From the publisher’s perspective, the obvious question is:
**Why exactly is everyone in this transaction getting paid except the people who made the material?**
It is not a ridiculous question.
But Washington has now effectively responded with another one:
**What if forcing AI companies to obtain permission for everything makes America worse at AI?**
And that is where copyright law suddenly becomes geopolitics.
Washington Says the Machine Is Doing Something New
The government’s brief argues that AI training is “extraordinarily” transformative.
That word matters.
American fair-use law does not simply ask whether copyrighted material was copied. It asks, among other things, what the new use does with the original work.
A photocopier making another copy of tomorrow’s newspaper is one thing.
A system analyzing millions of texts to learn statistical relationships between language, facts, concepts and expression is potentially another.
OpenAI and other AI companies have built much of their legal defense around that distinction.
Now the U.S. government is endorsing it.
Not absolutely.
Not for every imaginable AI use.
But strongly enough that the Justice Department told the court the United States has a significant interest in rejecting the theory that training itself generally violates copyright.
Why?
Because, according to the government, restricting model development on that basis could damage scientific advancement, American prosperity and national security.
That is a remarkable escalation.
Copyright holders are no longer arguing only against technology companies.
They are arguing against an emerging theory of national interest.
Your Copyright Has Met National Security
“National security” is one of those phrases that changes the temperature of a room.
A normal commercial dispute asks:
Who owns this?
Who used it?
Who owes whom?
A national-security dispute asks:
What happens if China gets there first?
And suddenly the balance changes.
The administration’s position fits neatly into its broader AI strategy.
At the G20 technology meeting in North Carolina, Commerce Secretary Howard Lutnick urged governments to embrace fair-use principles that allow AI companies room to train on protected material while still finding ways to protect creators.
At the same meeting, American officials pushed a generally lighter regulatory approach to AI, arguing that governments should avoid erecting barriers that slow innovation without addressing concrete harms.
Put those pieces together and the policy becomes obvious.
The United States does not merely want successful AI companies.
It wants AI development treated as strategic infrastructure.
And strategic infrastructure gets different political treatment from an ordinary software business.
The New York Times Has an Awkward Problem
The Times has spent years publishing stories about the power of technology companies.
It has reported extensively on privacy, misinformation, monopoly power, algorithmic influence, AI safety and the enormous political leverage of Silicon Valley.
Now it has found itself confronting a new version of that power.
Not simply:
**OpenAI is big.**
But:
**OpenAI may be strategically important to the United States.**
That is a much more difficult opponent.
A newspaper can argue with a corporation.
It can demand licensing fees.
It can sue.
It can ask a judge to enforce copyright law.
But once the government begins framing AI capability as a national-security asset, the newspaper is implicitly being asked to prove that its copyright claim should outweigh part of America’s technological competition with the rest of the world.
That is an extraordinary burden to place on copyright law.
It may also be exactly where the United States is heading.
Does “AI Needs It” Become a New Exception?
This is where everyone should become slightly uncomfortable.
The government may be correct that training large models is transformative.
Courts may ultimately decide that large-scale AI training falls within fair use.
There are serious legal arguments for that position.
But national-security rhetoric introduces a dangerous temptation.
Once a technology is declared strategically essential, almost every restriction on that technology can be described as a threat to competitiveness.
Licensing requirement?
China will win.
Copyright payment?
China will win.
Privacy restriction?
China will win.
Safety rule?
China will win.
Environmental limit?
China will win.
At some point “we need to compete” can become an all-purpose solvent for inconvenient law.
That does not mean the argument is wrong.
It means it deserves scrutiny precisely because it is so powerful.
The Creator’s Problem Is Real Too
AI companies sometimes talk about training data as though it arrived from nature.
It did not.
People wrote those books.
People recorded those songs.
People produced those photographs.
People paid reporters to spend months investigating stories.
The fact that an AI model processes those works in a technically novel way does not erase the economic question.
What happens if AI products become good enough to replace part of the market for the material that trained them?
Imagine a system that learns from thousands of paid news articles and then answers enough questions that users stop visiting those publications.
The AI company may say:
“We did not reproduce the articles. We learned from them.”
The publisher may answer:
“Wonderful. You learned how to replace us.”
That is not an easy dispute to solve with a slogan about innovation.
But Permission for Everything May Be Impossible
The opposite solution has its own absurdity.
Suppose every AI developer must negotiate permission individually with every copyright holder whose work might enter a training set.
Millions of authors.
Thousands of publishers.
Photographers.
Musicians.
Software developers.
Academic journals.
Film studios.
News agencies.
Websites that no longer exist.
Rights holders who cannot be identified.
Rights split across countries, companies, estates and licensing organizations.
The transaction costs alone could make large-scale model training practically impossible except for the richest companies on Earth.
Ironically, a rule intended to discipline Big Tech could create an even stronger monopoly for Big Tech.
OpenAI, Google, Meta and Microsoft might eventually afford giant licensing regimes.
A startup probably would not.
So copyright protection could end up protecting incumbents too.
There is no simple moral victory available here.
Europe Is Heading in the Other Direction
The transatlantic contrast is becoming difficult to miss.
European regulators are increasingly asking how publishers can prevent their content from being used in AI-generated search answers without sacrificing visibility in ordinary search.
The underlying principle is control:
**If my content feeds your AI product, I should have meaningful power to say no.**
Washington is moving toward a different emphasis:
**If requiring permission interferes with strategically important AI development, fair use should leave room for training.**
Neither position is completely irrational.
But they lead to very different internets.
Europe risks building a permission-heavy environment where AI development becomes slower and more legally expensive.
America risks creating an environment where creative work becomes raw industrial input simply because AI companies can process it at enormous scale.
One side fears exploitation.
The other fears losing the race.
Fair Use Was Never Designed for This Scale
This may be the real problem.
Fair use evolved through disputes involving human-scale acts of copying, criticism, parody, scholarship, news reporting and transformation.
Generative AI operates at a scale copyright law never anticipated.
A human researcher reads fifty books.
A model may ingest fifty million.
A journalist studies another newspaper.
A training system processes much of the accessible internet.
Scale changes the economics even if the underlying legal principle remains the same.
And courts are now being asked to take doctrines built for one technological era and stretch them around systems capable of absorbing culture at industrial scale.
No wonder the results are inconsistent.
The first judges to address the AI-training fair-use question have already reached differing conclusions.
This is not settled law.
Washington simply wants it settled in a particular direction.
The Government Is Not a Neutral Observer Anymore
That is perhaps the biggest development.
Until now, it was possible to describe the AI copyright wars as a conflict between creators and technology companies.
That description is becoming obsolete.
The United States government has now told a federal court that it has a strategic interest in how this dispute is resolved.
And that means the sides are changing.
On one side:
Publishers, authors, musicians and other rights holders arguing that their work cannot simply become free industrial feedstock.
On the other:
AI companies arguing that training is transformative.
And now behind them:
A federal government saying that American science, prosperity and national security may depend on allowing that transformation to continue.
The New York Times wants to protect its articles.
Washington says AI may need them.
The courts now have to decide whether “need” has anything to do with copyright law at all.