The US Justice Department Tells a Court That Training AI on Copyrighted Books Is Fair Use
In the consolidated New York Times case against OpenAI and Microsoft, the DOJ filed on the side of the AI companies and directly contradicted the US Copyright Office report whose author was fired last year.
The US Department of Justice filed a brief last week in the consolidated copyright case brought by The New York Times and other rights holders against OpenAI and Microsoft, and it came down on the side of the AI companies. Training a large language model on copyrighted text, the department argues, is fair use.
Fair use is the US legal doctrine that lets you use copyrighted material without permission in limited circumstances, such as quoting a book in a review. Whether it stretches to hoovering up millions of articles to train a model is the question the courts have been chewing on since the Times sued in late 2023, demanding billions in damages and the destruction of models trained on its work.
The DOJ’s argument turns on a line between training and output. During training, whole works get copied, but those copies are never published, and what the model produces afterwards “often if not always” lacks substantial similarity to the originals. Treating the two as one act, the department says, is legally wrong. Its analogy is Joan Didion, who as a teenager retyped Hemingway’s stories to work out how his sentences functioned. Under the plaintiffs’ logic, the DOJ suggests, Didion would have owed somebody money every time she later published.
What is behind this
The filing is remarkable mostly for who it contradicts. In 2025 the US Copyright Office published a report concluding that blanket fair use does not cover AI training at scale, arguing that models work from perfect copies and generate competing content far faster than any human. The DOJ now says that report carries no binding legal authority and misread the case law. Register of Copyrights Shira Perlmutter was fired shortly after the report came out and is currently challenging her dismissal, a fact the DOJ notes in a footnote.
The scale objection has not gone away. One person copying a book to learn from it is not the same as a company turning that content into a mass-market product that competes with the source. Courts elsewhere have landed differently: a German court ruled against Suno on training data and rejected the fair use defence outright. The department’s brief is an opinion, not a ruling, and the judge is free to ignore it.
What this means for you: If you write, draw, code, or make music for a living, this is the fight over whether your published work can be used to train the systems you compete with, and the US government has just told a court it can. If you only use AI tools, expect no immediate change. But how this case lands will shape what the models are allowed to learn from, and therefore what they can do. If you are in Europe, note that the EU rules and the German rulings point the other way, so the answer may end up depending on where you are.
Sources
Mistral Raises 3 Billion Euros, the Largest Round a European Tech Company Has Ever Closed
Samsung Electronics led a Series D that values the French lab at more than 21 billion euros. The pitch to investors was not a better chatbot but control: open weights, own infrastructure, and no vendor lock-in.
