Global AI Firms Purchase Used Korean Books to Train Models

<>

Artificial intelligence companies are purchasing used Korean books to train advanced models, facing a shortage of native data and the threat of internet pollution, koreadaily.com reported on October 11, 2026. Global developers are scouring international markets for physical texts to prevent model degradation while trying to capture the nuance of local languages.

Model Collapse and the Hunt for Physical Books

This finding underpins why major technology firms are focusing intensely on purchasing physical books. Joo Jae-gul, a professor at the KAIST Graduate School of AI, explained that as AI-generated content multiplies online, older books written by humans without machine assistance have gained critical importance. Joo added that AI systems have essentially finished learning all existing internet material.

Anthropic executives acknowledged the necessity of books in court filings submitted in January regarding copyright lawsuits from publishers. According to those documents, a co-founder stated that training models on books helps AI learn proper writing rather than mimicking a low-quality internet tone.

Korean Data Scarcity and Global Tech Demand

The targeting of Korean-language used books stems largely from the severe scarcity of digital Korean data. Data from Common Crawl, a non-profit organization that archives and releases web pages at scale, shows that Korean-language pages account for only 0.8% of the global internet.

In Spain, large quantities of secondhand nonfiction books in Catalan, a minority language, have similarly been sold off. Seo Yong-seok, a professor at the KAIST Munshul Graduate School of Future Strategy, emphasized that fluency and cultural context are different. Expressions containing Korean sentiment and experience can only be learned through Korean-language materials, Seo said.

Model Imperialism and the Korean Knowledge Economy

Another driving factor behind the acquisition of Korean books is the scale of the domestic AI market. Microsoft data published in January places South Korea second globally behind the United States in paid ChatGPT subscribers. Seo described the widespread purchasing of used books by tech giants as model imperialism, drawing a parallel to colonial trade where raw materials were cheaply extracted, processed abroad, and sold back as expensive finished goods.

Under this dynamic, knowledge accumulated by Korean society over the past century leaves the country cheaply as raw paper, returns as foreign AI models, and forces domestic consumers to pay for finished services. Seo warned that South Korea is shifting from a knowledge producer into a supplier of raw knowledge materials and a consumer of foreign AI services.

>