Anthropic Scores Major Fair Use Win in Authors’ Copyright Lawsuit

Training LLMs on protected works without authorization is fair use – at least according to a new federal court ruling in a copyright lawsuit filed against Anthropic by several authors.

Judge William Alsup just recently granted multiple (but not all) components of Anthropic’s summary judgement motion on the ever-important fair use question. As many know, AI giants are adamant that ingesting copyrighted materials to train their models is transformative and entirely lawful.

Unsurprisingly, the stance isn’t sitting right with professionals and rightsholders, who’ve fired off a number of related suits. Music publishers levied one of those (possibly settling) complaints, though it’s worth noting a key difference between the action and the (book) authors’ suit.

In short, the music publishers claim that Claude incorporated protected lyrics into its outputs, while the authors’ arguments center solely on training.

“Here, if the outputs seen by users had been infringing, Authors would have a different case,” Judge Alsup wrote in the summary judgement order. “And, if the outputs were ever to become infringing, Authors could bring such a case. But that is not this case.”

With the significant distinction out of the way, today’s summary judgement decision encompasses a few components, all with potentially far-reaching implications for the music space.

First, the judge expressed the belief that the actual process of training Claude on books (many of which having been purchased, stripped from the binding, and then scanned into a central library) fell under the fair use banner, notwithstanding the many copies made along the way.

“In short,” the court wrote, “the purpose and character of using copyrighted works to train LLMs to generate new text was quintessentially transformative. Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different.”

(To state the obvious, readers aspiring to become writers aren’t machines and certainly aren’t valued at nearly $62 billion.)

“Here, it is true that Anthropic could have used some other books or no books at all for training its LLMs — or so this order accepts the record shows for Authors,” the judge continued. “But Anthropic has presented a compelling explanation for why it was reasonably necessary to use them anyway.”

Perhaps equally pressing for the music world, where secondhand CDs are far from expensive or hard to find, Judge Alsup determined that the aforementioned physical-to-digital book-scanning process was, albeit for different reasons, also fair use.

“On the facts here,” he specified of Anthropic’s effort to assemble a so-called research library with millions of purchased books, “that format change itself added no new copies, eased storage and enabled searchability, and was not done for purposes trenching upon the copyright owner’s rightful interests — it was transformative.”

At the intersection of the library, subsequent copies, and training, “the further copies made therefrom for purposes of training LLMs were themselves transformative,” the court indicated.

All that said, Anthropic’s use of pirated books in its library – “Anthropic thereby pirated over seven million copies of books” – cannot constitute fair use, per Judge Alsup.

“We will have a trial on the pirated copies used to create Anthropic’s central library and the resulting damages, actual or statutory (including for willfulness). 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 wrote.