Landmark ruling says AI can read copyrighted books

AI models can also experience the joy of curling up with a new book, according to a decision filed late Monday by a federal judge, who ruled that Anthropic is allowed to train its chatbots on published books without the authors’ permission.

It marks the first-ever decision on whether using copyrighted materials for AI training is legal, but does come with some fine print.

• Namely, the AI startup is still on the hook for how they acquire those books.
• That’s because Judge William Alsup’s ruling distinguishes between legally obtained books, which he says are legal for AI to train on, and pirated books, which are not.

AI learns like humans do, judge says

“The court treats the AI as akin to a human learning from copyrighted material,” University of Miami law professor Christina Frohock told the Wall Street Journal. “It’s fair use if you and I pick up a book and read it and develop our own thoughts,” and the court concluded the same is true for AI systems, Frohock said.

It has a far-reaching impact: This week’s ruling strikes a blow to the hundreds of authors, artists, and publishers who have brought dozens of lawsuits against major AI companies—including Meta, Google, Midjourney, Stability AI, etc.—accusing them of illegally training their models on copyrighted work.

• While this week’s ruling doesn’t guarantee that other judges will follow the same lead, it does set a precedent making it more likely for courts to side with AI companies over creatives.

The dust won’t settle for a while. Analysts widely expect the Supreme Court to make a final decision on the matter at some point in the next few years, given the vast number of AI copyright cases working their way through the US legal system.