TLDR: Anthropic, the AI developer behind Claude, has agreed to a proposed $1.5 billion settlement in a class-action lawsuit filed by authors and publishers. The settlement addresses allegations of using approximately 500,000 pirated books to train its large language models, offering around $3,000 per work and requiring the destruction of unlawfully obtained files. This landmark agreement, though still subject to court approval, provides critical guidance on intellectual property rights and ethical data sourcing for the burgeoning AI industry.
In a pivotal development for the artificial intelligence industry, Anthropic, the company behind the Claude AI system, has reached a proposed $1.5 billion settlement in a class-action lawsuit concerning the use of copyrighted materials for AI training. This agreement, announced on September 17, 2025, is poised to establish a clearer framework for intellectual property in AI development, offering a significant roadmap for tech companies navigating the complex legal terrain of data acquisition.
The lawsuit, known as Bartz v. Anthropic (3:24-cv-05417 (N.D. Cal.)), was initiated in August 2024 by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, alongside a class of publishers. They alleged that Anthropic utilized pirated copies of approximately 500,000 books, sourced from online repositories such as Books3, LibGen, and Pirate Library Mirror, to train its Large Language Models (LLMs). The proposed settlement stipulates a compensation of approximately $3,000 per infringed work and mandates Anthropic to destroy all unlawfully obtained files.
Judge William Alsup, overseeing the case in the U.S. District Court for the Northern District of California, issued a crucial summary judgment ruling in June 2025. He distinguished between the use of lawfully acquired books for AI training, which he deemed ‘quintessentially transformative’ and protected by fair use, and the creation and retention of a ‘central library’ comprising pirated works, which he ruled was not transformative and constituted infringement. This nuanced decision set the stage for a trial on damages, leading to the current settlement negotiations.
While the $1.5 billion payout is substantial, representing one of the largest copyright recoveries in history, it also reflects a ‘massive discount’ for Anthropic. The company could have faced statutory damages up to $150,000 per work, potentially escalating its total exposure to over $1 trillion. Judge Alsup initially raised questions about the adequacy and notice procedures of the proposed agreement, even denying it as ‘nowhere close to complete,’ indicating that final terms may still evolve.
Also Read:
- Hollywood Giants Unite: Disney, Warner Bros. Discovery, and Universal Sue Generative AI App Hailuo for Mass Copyright Infringement
- American Arbitration Association Unveils AI Arbitrator for Construction Disputes
The settlement’s terms are particularly instructive for the AI sector. Notably, the agreement provides only a limited release covering Anthropic’s past training behavior, offering no license for future training. This means Anthropic remains exposed to potential lawsuits over AI-generated outputs that might infringe existing works. Experts emphasize that this structured approach underscores the financial and reputational risks associated with training AI systems on unlawfully obtained materials. Companies developing or deploying AI tools are now strongly advised to proactively ensure their data sourcing strategies withstand judicial and regulatory scrutiny, investing in legitimate data acquisition upfront to avoid the far greater costs of litigation and reputational damage.
