Key facts
- Publishers cannot always determine if their books are used for AI training.
- AI companies are acquiring books in bulk, with some denying AI training purposes but identifying data companies as final recipients.
- The Korean Publishers Society's appeal, "No Books, No AI," demands AI companies record and disclose training data sources and obtain rights holder permission.
- Over 240 publishers and organizations, along with 702 authors, translators, and illustrators, had joined the appeal as of Wednesday.
South Korean publishers are raising alarms over a surge in bulk book purchase requests, which they suspect are linked to artificial intelligence training. These requests have fueled concerns about intellectual property theft and a lack of compensation for creators. Publishing groups are investigating these suspicious orders and seeking a coordinated response, but face challenges in verifying the intended use of the books.
Kyobo Book Centre, a major bookstore chain, reported that since early this year, 10 overseas companies, including those from the U.S. and UK, inquired about purchasing one copy each of around 100,000 titles. While some denied an AI training purpose, they identified data-related companies as the final recipients, leading Kyobo to decline the requests. Yes24, another bookstore, also received a similar inquiry from a U.S. distributor.
Last week, the Literature Translation Institute of Korea informed authors whose books were identified on a compensation eligibility list associated with a U.S. copyright settlement involving Anthropic, an AI company facing a class-action lawsuit for allegedly downloading copyrighted works for model training.
The Korean Publishers Association is surveying industry players about these bulk orders, with the aim of forming a joint body to establish compensation arrangements and legislation. The Korean Publishers Society has issued an appeal, "No Books, No AI," demanding fair compensation and transparency from AI companies regarding the use of published works. Over 240 organizations and 702 individuals have signed the appeal, which calls for AI companies to record, disclose, and allow verification of training data sources and to obtain explicit permission from rights holders.