American AI firms suspected of buying thousands of used books in Japan to destroy them
Booming sales, but at what cost?
Image by VTT Studio | Shutterstock
- Used bookstores in Japan have seen online orders surge, with some sales rising fivefold since August.
- The bulk orders were traced to an Okayama logistics center and shipped to the US.
- Booksellers suspect AI companies scan and destroy specialized books to train artificial intelligence systems.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
Bookstores selling used books in Japan are seeing a significant surge in online orders. However, the initial excitement over the booming sales soon gave way to suspicion about where the books actually end up.
Since August, bookstores selling used books in Japan have been receiving large orders online.
Some booksellers say there were days when their sales increased fivefold, sometimes selling as many as 100 books a day.
This sudden interest in used books led to speculation that they’re actually being used to train AI and will be destroyed in the scanning process.
An investigation by journalists confirmed these suspicions, revealing that 50 tons of books were exported from Japan to the US, according to NTV Japan.
Buying books in bulk
Bookstores that sell their books on major online marketplaces appear to be the primary target, according to the report.
Industry insiders told the journalists that all orders come from multiple accounts across different stores. However, all are being shipped to a logistics center in Okayama Prefecture.
The data on sold books reveal that the orders include books on history, philosophy, politics, medicine, and law. It also includes readings on life and culture in the Edo period.
Bookstore representatives said that while the orders include all kinds of books, from comics to novels, most are for specialized titles.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
Getting AI training data at all costs
The Tokyo Used Book Association, an organization of used books in Japan, said it’s familiar with the situation, as discussions about these books being used for AI training already circulate in the industry.
Last year, it was revealed that Anthropic has been using print books to train Claude, thereafter destroying them.
Later, the company agreed to a $1.5 billion settlement after being sued by authors over illegally obtained books used to train Claude.
However, the court agreed that Anthropic infringed copyright by obtaining the books illegally, but not for using them to train AI.
The books used to train Anthropic’s Claude also included books from Japanese distributors.
More than 50 tons of “Japanese books” have been shipped to the US since last year, according to the news outlet.
Anthropic isn’t the only AI company looking for fresh data sources to train its models.
An AirTag was used to follow rare book shipments, which led to Amazon's AI training facility, where books were torn apart to prepare them for scanning, and their contents were used to train AI.
