Anthropic Settles Copyright Suit for $1.5 Billion
The AI firm agreed to a massive settlement after using millions of pirated digital books to train its Claude model.
Updated on Oct. 16, 2026 in Artificial Intelligence
Live Poll
Should AI companies be required to compensate creators when using copyrighted material for model training?
Anthropic has agreed to a $1.5 billion settlement following a class-action lawsuit over its unauthorized use of copyrighted books. The company downloaded over 7 million digital copies from illegal piracy sites between 2021 and 2022 to train its Claude AI model.
Why it matters
A U.S. court ruled that downloading and storing pirated content for AI model development does not constitute fair use. This settlement resolves legal action involving nearly half a million individual works.
The settlement covers 482,460 works, with Anthropic paying approximately $3,000 per book. The company originally obtained 5 million copies from Library Genesis in 2021 and 2 million from Pirate Library Mirror in 2022.
The players
Anthropic
This American artificial intelligence startup develops the Claude model and is currently transitioning to licensed training materials.
Andrea Bartz
She is an author who filed the class-action lawsuit against the AI developer in 2024.
The details
The lawsuit, initiated by author Andrea Bartz in 2024, highlighted the company's reliance on illicit digital archives. In response to the ruling, Anthropic shifted its training strategy in 2024 to focus on purchasing and digitizing paper books in bulk.
Timeline
In 2021, Anthropic downloaded 5 million copies from Library Genesis.
In 2022, the company downloaded 2 million copies from Pirate Library Mirror.
In 2024, authors filed a class-action lawsuit against Anthropic.
In 2024, Anthropic transitioned to purchasing and digitizing paper books.
On October 16, 2026, Anthropic published the copyright settlement book list.
The Tech Race
This settlement highlights a major shift in the industry as AI developers move away from unlicensed web scraping. Companies are increasingly replacing legacy data-gathering techniques with regulated, direct-purchase models to avoid similar litigation.
The transition to licensed content may result in slower training cycles or higher costs for future AI service iterations. Users should expect AI companies to increasingly pivot toward verified, legal data sources for their model development.
The takeaway
This case underscores the financial risks companies face when using unauthorized data to train high-level AI models. Developers must now weigh the costs of legal settlements against the expense of acquiring legal, digitized training materials.
Further reading
For more context on the evolving legal standards for machine learning, visit the Artificial Intelligence section.
Source note: This article includes information reported by 조선일보.
Live Poll
Should AI companies be required to compensate creators when using copyrighted material for model training?
