Transformative.
That’s how a federal court characterized Amazon-backed Anthropic‘s use of millions of books across the web to teach its artificial intelligence system. It’s the first decision to consider the issue and will serve as a template for other courts overseeing similar cases. And studios, now that some have entered the fight over the industry-defining technology, should be uneasy about the ruling.
The thrust of these cases will be decided by one question: Are AI companies covered by fair use, the legal doctrine in intellectual property law that allows creators to build upon copyrighted works without a license? On that issue, a court found that Anthropic is on solid legal ground, at least with respect to training.
The technology is “among the most transformative many of us will see in our lifetimes,” wrote U.S. District Judge William Alsup.
Still, Anthropic will face a trial over illegally downloading seven millions books to create a library that was used for training. That it later purchased copies of those books it stole off the internet earlier to cover its tracks doesn’t absolve it of liability, the court concluded. The company faces potential damages of hundreds of millions of dollars stemming from the decision that could lead to Disney and Universal getting a similar payout depending on what they unearth in discovery over how Midjourney allegedly obtained copies of thousands of films that were repurposed to teach its image generator.
Last year, authors filed a lawsuit against Anthropic accusing it of illegally downloading and copying their books to power its AI chatbot Claude. The company chose not to move to dismiss the complaint and instead skipped straight for a decision on fair use.
In the ruling, the court found that authors don’t have the right to exclude Anthropic from using their works to train its technology, much in the same way they don’t have the right to exclude any person from reading their books to learn how to write.
“Everyone reads texts, too, then writes new texts,” reads the order. “They may need to pay for getting their hands on a text in the first instance. But to make anyone pay specifically for the use of a book each time they read it, each time they recall it from memory, each time they later draw upon it when writing new things in new ways would be unthinkable.”
If someone were to read all the modern-day classics, memorize them and emulate a blend of their writing, that wouldn’t constitute copyright infringement, the court concluded. Like any reader who wants to be a writer, Anthropic’s technology draws upon works not to replicate or supplant them but to create something entirely different, according to the order.
Those aren’t the findings that Disney or Universal — both of whom are suing Midjourney for copyright infringement — wanted or expected. For them, there’s reason to worry that Alsup’s analysis will shape the way in which the judge overseeing their case considers undermining development of a technology that was found by another court to be revolutionary (or something close to it). More broadly, it could be found that AI video generators, like Sora, are simply distilling every movie ever made to create completely new works.
“This Anthropic decision will likely be cited by all creators of AI models to support the argument that fair use applies to the use of massive datasets to train foundational models,” says Daniel Barsky, an intellectual property lawyer at Holland & Knight.
Important to note: The authors didn’t allege that responses generated by Anthropic infringed upon their works. And if they had, they would’ve lost that argument under the court’s finding that guardrails are in place to ensure that no infringing ever reached users. Alsup compared it to Google imposing limits on how many snippets of text from any one book could be seen by a user through its Google Book service, preventing its search function from being misused as a way to access full books for free.
“Here, if the outputs seen by users had been infringing, Authors would have a different case,” Alsup writes. “And, if the outputs were ever to become infringing, Authors could bring such a case. But that is not this case.”
But that could be the case for Midjourney, which returns nearly exact replicas of frames from films in some instances.
When prompted with “Thanos Infinity War,” Midjourney — an AI program that translates text into hyper-realistic graphics — replied with an image of the purple-skinned villain in a frame that appears to be taken from the Marvel movie or promotional materials, with few to no alterations made. A shot of Tom Cruise in the cockpit of a fighter jet, from Top Gun: Maverick, is produced when the tool was asked for a frame from the film. The chatbots can seemingly replicate almost any animation style, generating startlingly accurate characters from titles ranging from DreamWorks’ Shrek to Pixar’s Ratatouille to Warner Bros.’ The Lego Movie.
“The fact that Midjourney generates copies and derivatives of” films from Disney and Universal proves that the company, without their knowledge or permission, “copied plaintiffs’ copyrighted works to train and develop” its technology, states the complaint.
Also at play: The possibility that Midjourney pirated the studios’ movies. In the June 23 ruling, Alsup found that Anthropic illegally downloading seven million books to build a library to be used for training isn’t covered by fair use. He said that the company could’ve instead paid for the copies. Such piracy, the court concluded, is “inherently, irredeemably infringing.”
With statutory damages for willful copyright infringement reaching up to $150,000 per work, massive payouts are a possibility.
