Anthropic AI — Illegal Downloads of Han Kang’s Books and 100,000 Korean Volumes Purchased

Last year, it was confirmed that at least 100,000 Korean books were exported to U.S. artificial intelligence (AI) startup Anthropic AI for AI training purposes. Separately, Anthropic illegally downloaded at least 22 books by prominent Korean authors, including Han Kang, Hwang Sok-yong, Park Wan-suh, and Kim Ae-ran, to train its AI models. Following court compensation rulings, the company has reportedly offered compensation to author Han Kang and others.

Internal meeting records from Anthropic, submitted during the ‘Bartz v. Anthropic’ lawsuit finalized in July of last year, detail the progress of a contract with South Korean book distributor Woongjin Booksen dated November 27, 2024. The documents indicate that the contract terms were internally approved, agreements on discount rates were finalized, and a revised contract was received from Woongjin Booksen. Although the exact price per book and total purchase scale were kept private, the purchase scale was noted as ‘0.Xm (million),’ confirming a volume of at least 100,000 books.

Woongjin Booksen confirmed to the JoongAng Ilbo that it exported books to Anthropic last year through a direct contract without any intermediaries. The distributor explained that the transaction was initially understood as a standard book supply deal, adding that following recent controversies regarding book scanning and shredding, all transactions with AI companies or suspicious buyers have been completely suspended. The specific export volume remains undisclosed.

In addition to physical book imports, court proceedings in the United States revealed that Anthropic illegally downloaded electronic books for AI training. The list of illegally downloaded materials includes works by Korean authors such as Han Kang, Hwang Sok-yong, Shin Kyung-sook, Kim Young-ha, Hwang Sun-won, Kim Un-su, Kim Ae-ran, Bae Su-ah, and Kim Sum, spanning 22 titles across 15 authors.