09/03/2026
The US government just walked into the biggest copyright fight in tech and stood next to OpenAI.
On Tuesday the Justice Department told a federal court that training AI models on copyrighted books, articles, and songs is fair use. Its stated reason: ruling the other way would be a national security risk.
This is the case The New York Times filed in 2023, accusing OpenAI and Microsoft of using millions of its articles without permission to build ChatGPT. It is one of dozens of lawsuits like it, brought by authors, publishers, music labels, and newsrooms. Until this week the government had stayed out of all of them.
Now it is on the record. The DOJ brief calls AI training "extraordinarily transformative" and argues that forcing companies to license their training data would hand foreign competitors an edge, slow scientific research, and lock the technology inside the few companies large enough to afford the licensing bills. The Times responded that the administration is siding with "a handful of trillion-dollar AI companies at the expense of the countless American creators whose work they stole."
Here is what it means for you.
The brief settles nothing. It carries advisory weight, not legal weight, and the judge can read it and set it aside. The first two judges to rule on this question last year landed on opposite sides, so the law is still open.
But the direction is clearer than it was a week ago. Every business built on top of these models has carried a quiet question for two years: what happens to my tools, my workflows, my product, if a court rules the training data was stolen. That question is still unanswered. It just got less likely to end in the worst case.
The other side of it is just as real. If you produce content, writing, images, music, research, the same brief argues you have no claim to payment when a model trains on your work. Licensing leverage is what creators have right now, and the government is on record trying to weaken it.
So two things are true at once. If you build with AI, the ground under you looks firmer this week. If you sell content, it looks softer. Most operators sit on both sides of that line.
The practical move does not change either way. Know where your AI vendors got their training data, and ask them in plain terms. Read the indemnification language in your contracts, the part that says who pays if a model output triggers a lawsuit. The businesses that get hurt in the next two years will be the ones who treated this as settled because a government brief said it was.
If a court does rule that training on copyrighted work is fair use, does that change what you are willing to build on top of, or were you never worried about it to begin with?
