According to Fortune, Anthropic, a leading artificial intelligence (AI) laboratory, faces a class-action lawsuit that could result in multi-billion dollar copyright damages due to its alleged use of pirated books from shadow libraries like LibGen and PiLiMi to train its large language model, Claude. The lawsuit claims that Anthropic downloaded millions of copyrighted works with the intention of building a “central library” of digital books, aiming to include “all the books in the world” and preserve them indefinitely.
The plaintiffs in the case, including authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, contend that a substantial portion of these works were obtained from piracy websites, directly violating copyright law. The legal battle centers on the methods Anthropic employed to acquire its training data.
Judge William Alsup, presiding over the case, has made a significant ruling that training AI models on lawfully acquired books qualifies as “fair use.” This decision, seen as a major win for the AI sector, suggests that AI companies do not require a license from copyright holders for such training. However, a critical distinction was drawn regarding the acquisition of materials. The unresolved issue remains how Anthropic obtained and stored the copyrighted books, particularly those from unauthorized sources.
Judge Alsup indicated that a separate trial would be held specifically to address the “pirated copies” and the “resulting damages.” Luke McDonagh, an associate professor of law at LSE, explained the nuance, stating, “The judge seems to be suggesting that if you had bought a million books from Amazon in digital form, then you could do the training, and that would be legal, but it’s the downloading from the pirate website that is the problem, because there’s two things, there’s that acquiring of the copy, and then the use of the copy.”
Ed Lee, a law professor at Santa Clara, commented in a blog post that the ruling could expose Anthropic to “at least the potential for business-ending liability.” While plaintiffs are unlikely to prove direct financial harm, such as lost sales, they are expected to pursue statutory damages. These damages can range from $750 to $150,000 per work, with the higher end reserved for cases where infringement is deemed willful.
The potential financial implications for Anthropic are substantial. While the exact number of works included in the class action and whether willful infringement will be proven remain uncertain, legal experts estimate potential damages could range from hundreds of millions to tens of billions of dollars. Even at the lower end, Lee suggests damages between $1 billion and $3 billion are possible if only 100,000 works are included in the class action. This figure rivals the largest copyright damage awards on record and could significantly surpass Anthropic’s current annual revenue of $4 billion. Lee further estimated that if a jury finds the company willfully pirated 6 million copyrighted books, the liability could skyrocket to as much as $1.05 trillion.
Anthropic has not publicly commented on the latest developments but previously stated it “respectfully disagrees” with Judge Alsup’s decision and is exploring its options, including an appeal or a settlement.
The trial, which is the first certified class action against an AI company concerning the use of copyrighted materials, is scheduled for December 1. Its outcome could establish a precedent for similar ongoing cases, such as the high-profile dispute between OpenAI and various authors and publishers.
A divergence in judicial opinions exists regarding the acquisition of copyrighted works from shadow sites. In a separate copyright case against Meta, Judge Vince Chhabria argued that the transformative purpose of AI use could legitimize initial unauthorized downloading. McDonagh noted that Judge Chhabria’s ruling suggested that the positive, transformative use of the works could “correct” the initial problematic acquisition. In contrast, Judge Alsup views the downloading of books from unauthorized shadow libraries as “inherently wrong,” implying that even if AI training is considered fair use, the initial acquisition of works was illegitimate and requires compensation.
The judges also differed on whether AI-generated outputs compete with the original copyrighted works used in their training data. Judge Chhabria acknowledged that such competition could weaken a fair use defense but found insufficient evidence of market harm in the Meta case. Judge Alsup concluded that generative AI outputs do not compete with the original works at all.
The current U.S. administration is advocating for broad fair use protections for AI companies using copyrighted materials, a strategic move aimed at maintaining U.S. leadership in artificial intelligence.
Despite the significant potential damages, McDonagh expressed skepticism that the lawsuit would lead to Anthropic’s bankruptcy. He suggested that the U.S. Supreme Court, currently perceived as aligned with the Trump administration’s agenda, would likely prevent a ruling that could dismantle an AI company. Additionally, courts generally avoid issuing rulings that could result in bankruptcy unless there is a strong legal basis and the action is deemed absolutely necessary. Courts often consider the potential impact on the company and its stakeholders when issuing rulings that could lead to liquidation.
McDonagh concluded, “Anthropic is now valued, depending on different estimates, between $60 and $100 billion. So paying a couple of billion to the authors would by no means bankrupt the organization.”
