裁判官がAnthropicの15億ドル書籍海賊版著作権侵害和解を承認Anthropic’s $1.5 billion book piracy settlement approved by judge

A federal judge has granted final approval to Anthropic's $1.5 billion settlement with a group of authors who accused the company of using pirated books to train its Claude family of large language models. The decision ranks among the largest copyright resolutions the technology industry has seen, and it arrives while courts, creators, and AI developers are still working out how longstanding intellectual property law applies to generative systems.

The case, filed as a class action by authors including Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, focused less on the act of training itself than on how Anthropic gathered the underlying text. Plaintiffs alleged that the company downloaded large collections of books from pirate repositories, including sites such as Library Genesis, rather than licensing or purchasing them.

The litigation produced a closely watched ruling from US District Judge William Alsup that separated two distinct questions. He indicated that training a model on lawfully obtained books could constitute fair use, calling the transformation of that text into a model "exceedingly transformative." At the same time, he concluded that downloading and storing pirated copies to build a permanent digital library fell outside fair use and could constitute infringement, leaving Anthropic exposed to substantial damages.

That split is important because it isolates the piracy question from the broader, unresolved debate over whether AI training is inherently lawful. With a large number of works in play, statutory damages could in theory have reached far beyond the settlement figure had the dispute proceeded to a damages trial, an outcome that appears to have encouraged both sides to reach an agreement. Under the terms, Anthropic is expected to pay roughly $3,000 per work across an estimated 500,000 titles and to destroy the disputed datasets, while not conceding liability.

For authors and publishers, the amount establishes a concrete reference point for what unlicensed use of copyrighted material can cost, even if a settlement does not carry the binding precedential weight of a court judgment. Anthropic, backed by major investors including Amazon and Google and known for positioning itself as a safety-focused AI developer, is likely to treat the payment as a cost of clearing legal uncertainty as it continues to expand Claude and its enterprise business.

The settlement sits within a wider wave of litigation over training data. OpenAI and Microsoft are defending a high-profile suit brought by The New York Times, while Meta, Stability AI, and other developers face comparable claims from writers, artists, and news organizations. In a separate case involving Meta, a different judge reached conclusions favorable to the company, though on narrow grounds, underscoring how unsettled and fact-specific this area of law remains.

Running parallel to the courtroom fights is a growing market for licensed content. Companies including OpenAI have signed agreements with publishers such as News Corp and Axel Springer, and Anthropic itself has pursued content arrangements, signaling an industry gradually moving toward paid data partnerships as one way to reduce legal risk. The distinction