Anthropic is facing a class-action lawsuit that threatens to radically change how AI companies obtain training data for their large language models. The case centers on allegations that the company downloaded millions of copyrighted works from shadow libraries, including LibGen and PiLiMi, to develop the Claude model, now one of the most widespread generative AI models worldwide. According to AI Magazine, a federal judge, William Alsup, recently issued a ruling creating uncertainty in the AI sector. He decided that training AI models on legally purchased books constitutes fair use, eliminating licensing agreement requirements with copyright holders. However, Alsup made a crucial distinction regarding how companies acquire these materials.

Luke McDonagh, co-law professor at the London School of Economics, said, “The judge seems to suggest that if you bought a million books from Amazon digitally, you can train on them legally, but downloading from piracy sites is the problem.” The case has become a class action, allowing any author whose work was used without permission to join, expanding potential damages from a limited number of plaintiffs to thousands. The financial consequences for Anthropic could be catastrophic. Ed Lee, law professor at Santa Clara University, notes the ruling could expose Anthropic to liabilities that could end its business. Legal damages range from $750 to $150,000 per work, with the maximum applied if infringement is deemed willful. Ed estimates Anthropic could face damages between $1 billion and $3 billion if the class includes only 100,000 works.

In the worst-case scenario, the company could face fines up to $1.05 trillion if a jury finds Anthropic deliberately pirated 6 million copyrighted books, dwarfing the company’s valuation and current resources. The burden of proof may shift to Anthropic to demonstrate it obtained training materials legally. Richard Johnson, co-founder and COO of Data Guardians Network, explains, “Suddenly, Anthropic may be asked to prove it did not use pirated data—or at least obtained it legally. If not possible, it faces a legal and reputational crisis that could last a long time.” The class action status significantly increases financial risks. Richard says, “This turns a few plaintiffs into thousands, meaning damages could reach billions.” This is the first certified class-action lawsuit against an AI company over copyrighted material use, with trial set for December 1, 2025. Legal experts expect this precedent to spark similar lawsuits across the AI sector.

Richard believes the industry faces an accountability moment: “For years, everyone knew AI companies took shortcuts and that stealing copyrighted materials without user consent was common, but the race to outpace competitors made it an ‘acceptable risk.'” The implications extend beyond Anthropic to the broader AI ecosystem. Companies that developed models using similar data collection methods may face similar legal challenges. Richard says, “If the court rules training on pirated data is not fair use, it could invalidate a vast number of current models overnight.” The current political environment may influence the final ruling. Luke notes the Trump administration would likely resist any ruling that could “fundamentally destroy an AI company,” given rising international competition between the US and China in AI supremacy. Thus, the political dimension complicates what might seem a straightforward copyright case and could affect regulatory approaches and appeals.

This case has already changed industry practices toward more cautious data acquisition methods. Richard explains, “That’s why ambitious companies seek pioneering data from ethical sources. Once this precedent is set, rebuilding datasets will be arduous and more expensive.” Public opinion appears to favor content creators over AI companies in these disputes, potentially influencing jury decisions and broader regulatory approaches. The outcome is likely to establish legal frameworks defining permissible data collection practices for AI training across the sector. Richard expects widespread legal actions following this case: “Many more lawsuits will be filed in the next 12-18 months. This litigation will force the AI sector to radically reconsider—not only compliance but also data acquisition ethics.”