Federal court says copyrighted books are fair use for AI training

Washington Post

4 hours ago

A federal judge this week ruled that artificial intelligence company Anthropic did not break the law when it used copyrighted books to train its chatbot, Claude, without the consent of the texts' authors or publishers — but he ordered the company to go to trial for allegedly using pirated versions of the books.
The decision, made Monday by Judge William Alsup of the U.S. District Court for the Northern District of California, represents a win for AI companies, which have battled copyright lawsuits from writers and news organizations for using their work to train AI systems.
Alsup said Anthropic's use of the books to train its large language models, was like an aspiring writer who reads copyrighted texts 'not to race ahead and replicate or supplant' those works, 'but to turn a hard corner and create something different.'
His ruling was on a lawsuit filed against Anthropic last year by three authors — Andrea Bartz, Charles Graeber and Kirk Wallace Johnson — who alleged that the company used their work without their consent to train AI systems in what amounted to 'largescale theft.'
But Alsup ordered Anthropic to face trial for the accusation that it knowingly obtained copies of more than 7 million books from piracy websites, although the company later paid to purchase copies of some books.
Alsup said he doubted that 'any accused infringer could ever meet its burden of explaining why downloading source copies from pirate sites that it could have purchased or otherwise accessed lawfully was itself reasonably necessary to any subsequent fair use.'
'That Anthropic later bought a copy of a book it earlier stole off the internet will not absolve it of liability for the theft but it may affect the extent of statutory damages,' he added.
In a statement, Anthropic said it was pleased that the court recognized that using published works to train LLMs was consistent with copyright laws 'in enabling creativity and fostering scientific progress.'
But the company said it disagrees with the decision to hold a trial for its 'acquisition of a subset of books and how they were used,' in apparent reference to the piracy allegations. 'We remain confident in our overall case, and are evaluating all options,' it said.
In their lawsuit, the authors said the actions of Anthropic have made 'a mockery of its lofty goals.' The company was founded in 2021 by a group that included OpenAI's former vice president of research Dario Amodei with goals that included 'research into increasing the safety of AI systems.'
Bartz and Johnson did not reply to requests for comment. Graeber declined to comment.
After concerns arose within the company about using pirated books, Anthropic hired former Google Books executive Tom Turvey to obtain 'all the books in the world' while also avoiding as many legal issues as possible, according to court documents.
Turvey and his team could have sought to reach commercial agreements with publishers to license the books to train its AI systems, Alsup noted, but they instead purchased millions of print books from retailers, many of them in used condition, then scanned them into digital form.
The company could have also hired staff writers and engineers to create good original writing to train AI models. But that would have 'required spending more,' Alsup noted.