Anthropic has agreed to a $1.5 billion settlement with authors in a landmark copyright case, marking one of the first and largest legal payouts of the AI era.
The AI startup agreed to pay authors around $3,000 per book for roughly 500,000 works, after it was accused of downloading millions of pirated texts from shadow libraries to train its large language model, Claude. As part of the deal, Anthropic will also destroy data it was accused of illegally acquiring.
The fast-growing AI startup announced earlier this week that it had just raised an additional $13 billion in new venture capital funding in a deal that valued the company at $183 billion. It has also said that it is currently on pace to generate at least $5 billion in revenues over the next 12 months. The settlement would amounts to nearly a third of that figure or more than a tenth of the new funding it just received.
While the settlement does not establish a legal precedent, experts said it will likely serve as an anchor figure for the amount other major AI companies will need to pay if they hope to settle similar copyright infringement lawsuits. For instance, a number of authors are suing Meta for using their books without permission. As part of that lawsuit, Meta was forced to disclose internal company emails that suggest it knowingly used a library of pirated books called LibGen—which is one of the same libraries that Anthropic used. OpenAI and its partner Microsoft are also facing a number of copyright infringement cases, including one filed by the Author’s Guild.
Aparna Sridhar, deputy general counsel at Anthropic, told Fortune in a statement: “In June, the District Court issued a landmark ruling on AI development and copyright law, finding that Anthropic’s approach to training AI models constitutes fair use. Today’s settlement, if approved, will resolve the plaintiffs’ remaining legacy claims. We remain committed to developing safe AI systems that help people and organizations extend their capabilities, advance scientific discovery, and solve complex problems.”
A lawyer for the authors who sued Anthropic said the settlement would have far-reaching impacts.
“This landmark settlement far surpasses any other known copyright recovery. It is the first of its kind in the AI era. It will provide meaningful compensation for each class work and sets a precedent requiring AI companies to pay copyright owners,” Justin Nelson, partner with Susman Godfrey LLP and co-lead plaintiffs’ counsel on Bartz et al. v. Anthropic PBC, said in a statement. “This settlement sends a powerful message to AI companies and creators alike that taking copyrighted works from these pirate websites is wrong.”
The case, which was originally set to go to trial in December, could have exposed Anthropic to damages of up to $1 trillion if the court found that the company willfully violated copyright law. Santa Clara law professor Ed Lee said could that if Anthropic lost the trial, it could have “at least the potential for business-ending liability.” Anthropic essentially concurred with Lee’s conclusion, writing in a court filing that it felt “inordinate pressure” to settle the case given the size of the potential damages.
The jeopardy Anthropic faced hinged on the means it had used to obtain the copyrighted books, rather than the fact that they had used the books to train AI without the explicit permission of the copyright holders. In July, U.S. District Court Judge William Alsup, ruled that using copyrighted books to create an AI model constituted “fair use” for which no specific license was required. But Alsup then focused on the allegation that Anthropic had used digital libraries of pirated books for at least some of the data it fed its AI models, rather than purchasing copies of the books legally. The judge suggested in a decision allowing the case to go to trial that he was inclined to view this as copyright infringement no matter what Anthropic did with the pirated libraries.
By settling the case, Anthropic has sidestepped an existential risk to its business. However, the settlement is significantly higher than some legal experts were predicting. The motion is now seeking preliminary approval of what’s claimed to be “the largest publicly reported copyright recovery in history.”
James Grimmelmann, a law professor at Cornell Law School and Cornell Tech, called it a “modest settlement.”
“It doesn’t try to resolve all of the copyright issues around generative AI. Instead, it’s focused on what Judge Alsup thought was the one egregiously wrongful thing that Anthropic did: download books in bulk from shadow libraries rather than buying copies and scanning them itself. The payment is substantial, but not so big as to threaten Anthropic’s viability or competitive position,” he told Fortune.
He said that the settlement helps establish that AI companies need to acquire their training data legitimately, but does not answer other copyright questions facing AI companies, such as what they need to do to prevent their generative AI models from producing outputs that infringe copyright. In several cases still pending against AI companies—including a case The New York Times has filed against OpenAI and a case that movie studio Warner Brothers filed just this week against Midjourney, a firm that makes AI that can generate images and videos—the copyright holders allege the AI models produced outputs that were identical or substantially similar to copyrighted works.
“The recent Warner Bros. suit against Midjourney, for example, focuses on how Midjourney can be used to produce images of DC superheroes and other copyrighted characters,” Grimmelmann said.
While legal experts say the amount is manageable for a firm the size of Anthropic, Luke McDonagh, an associate professor of law at LSE, said the case may have a downstream impact on smaller AI companies if it does set a business precedent for similar claims.
“The figure of $1.5 billion, as the overall amount of the settlement, indicates the kind of level that could resolve some of the other AI copyright cases. It could also point the way forward for licensing of copyright works for AI training,” he told Fortune. “This kind of sum—$3,000 per work—is manageable for a firm valued as highly as Anthropic and the other large AI firms. It may be less so for smaller firms.”
A business precedent for other AI firms
Cecilia Ziniti, a lawyer and founder of legal AI company GC AI, said the settlement was a “Napster to iTunes” moment for AI.
“This settlement marks the beginning of a necessary evolution toward a legitimate, market-based licensing scheme for training data,” she said. She added the settlement could mark the “start of a more mature, sustainable ecosystem where creators are compensated, much like how the music industry adapted to digital distribution.”
Ziniti also noted the size of the settlement may force the rest of the industry to get more serious about licensing copyrighted works.
“The argument that it’s too difficult to track and pay for training data is a red herring because we have enough deals at this point to show it can be done,” she said, pointing to deals that news publications, including Axel Springer and Vox, have entered into with OpenAI. “This settlement will push other AI companies to the negotiating table and accelerate the creation of a true marketplace for data, likely involving API authentications and revenue-sharing models.”
