A U.S. judge has ruled that training AI systems using copyrighted books can qualify as “transformative” under fair use law – but AI firm Anthropic must still face trial over how it obtained the works.

The case was brought by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused Anthropic of using pirated copies of their books to train its Claude AI model and profit from their work.

Judge William Alsup said the AI firm’s use of the books was “exceedingly transformative,” comparing the training process to how an aspiring writer might learn by reading. He noted the authors did not claim that Claude generated knockoff versions of their books – a factor that could have changed the outcome.

Still, Alsup denied Anthropic’s motion to dismiss the case, citing evidence that the company stored over 7 million pirated books in a “central library,” potentially violating the authors’ rights.

Anthropic, backed by Amazon and Alphabet, could face damages of up to $150,000 per work if found liable. In a statement, the company said it welcomed the fair use recognition but disagreed with the decision to go to trial and is reviewing its legal options.

The lawsuit is one of several testing how copyright laws apply to large language models. Similar cases are unfolding over the use of articles, music, and visual media. Some AI companies have started licensing original content to avoid legal challenges.