RDPLINE

A 100,000-Title Order: AI Wants Korean Content

2026. 10. 22

A 100,000-Title Order: AI Wants Korean Content

Several overseas companies wanted to buy 100,000 different Korean books, one copy each. They never said why.

According to a JoongAng Ilbo report on October 15, 2026, Kyobo Book Centre, Korea's largest bookstore, had been receiving bulk purchase inquiries from companies based in the US, the UK and elsewhere since early this year. The requests covered roughly 100,000 Korean-language titles regardless of subject, with one copy per title. The buyers declined to name their end customer, citing confidentiality.

On the same day, the Seoul Economic Daily reported that at least 15 overseas editions of works by Korean authors, including Han Kang, Hwang Sok-yong and Shin Kyung-sook, were on the settlement list in a US copyright case against Anthropic. The same company had also signed a book purchasing agreement with Woongjin Booxen, a major Korean book distributor.

The two cases are different in nature. The Kyobo orders are suspected, not confirmed, to be for AI training, while the Anthropic case involves both pirated copies and legitimate purchases. But together they point to one fact: Korean content is already on the data acquisition list of global AI companies.

For any organization that owns content, this raises a practical question.

If AI companies want to use your content, what uses will you allow, how far, and on what terms?

1. Selling a book is not the same as licensing it for AI training

The reports involve three different things: downloading pirated copies, buying physical books, and licensing content for AI training.

Acquiring pirated copies is clearly different from buying books legitimately. At the same time, a legitimate purchase does not automatically permit every form of digital copying and AI use.

Under Korean copyright law, as under the first-sale doctrine in the US, the right to control distribution is exhausted once a physical copy has been lawfully sold. That rule governs how the book can be resold. Whether the book can be digitized and used to train AI is a separate question.

This is where content owners should start. Check what rights your existing publishing and distribution contracts actually cover, and who holds the authority to permit AI training or use in search services.

2. The copyright dispute showed why acquisition matters

How Anthropic sourced its books became a central issue in Bartz v. Anthropic in the United States.

The court treated the use of books for AI training separately from the act of acquiring and storing pirated copies. The company later agreed to pay around 1.5 billion US dollars to settle claims related to the downloaded books.

That figure should not be read as the market price of content for AI training. A settlement in a copyright dispute and a fee for a legitimate data license are fundamentally different things.

What the case does show is how much provenance and rights clearance now matter when AI companies acquire data. The same applies to suppliers. To have a concrete conversation about a deal, you need to be able to explain what rights you can grant and under what conditions.

3. Demand for Korean content is already here

According to the Seoul Economic Daily, Anthropic's internal documents noted plans to order books from Taiwan, Japan and Korea, following the US and the UK. Woongjin confirmed the contract.

Kyobo's case fits the same pattern. Around 10 companies made inquiries, most of them from the US. Kyobo halted the talks because it could not rule out AI training use, and called on publishers and distributors to share information and agree on common principles.

A buying pattern that values the number of titles over the number of copies, and deals where no one knows who will use the content or for what. That is exactly why content owners need to set their supply terms first.

This does not mean every piece of Korean content has a buyer. But it is significant that Korean-language content is now a real acquisition target for global AI companies.

It is also a good moment to look at your own archive with fresh eyes.

Does it hold deep expertise in a specific field? Does it cover a long period of time? Is new material produced on a regular basis? Are the source and date of each item clear, and do you hold the necessary usage rights?

Once these characteristics and rights are organized, you can assess which buyers your data fits and how it could be supplied.

4. Rules and contract frameworks are taking shape

How copyright applies to AI training depends on the jurisdiction and the facts of each case. One ruling cannot be applied to all content.

In Korea, the Ministry of Culture, Sports and Tourism published guidelines in February 2026 on fair use of copyrighted works for generative AI training. The guidelines explain that fair use should be assessed by weighing the purpose and character of the use, the type of work, the amount used, and the effect on the market.

On October 8, the guidelines were also reported to Korea's National Intellectual Property Committee, and the government announced plans to release standard contracts for transferring and licensing copyrighted works for AI training by the end of the year.

In the EU, providers of general-purpose AI models are required to respect rights reservations made by copyright holders.

What content owners can do now is clarify the rights status of their materials and the conditions under which they can be used. With this in place, you have a basis to negotiate scope and price when a real offer arrives.

5. Four steps to prepare for data supply

Owning content is not the same as being ready to sell it. You need to define what buyers can actually use and what you are able to permit.

Step 1. Inventory your content and clarify rights

List what you hold and how much of it, and identify the rights holder for each item.

Contributed articles, co-authored works, translations, photos and illustrations may involve separate rights. Also distinguish between the right to store or publish content yourself and the right to let a third party use it for AI.

Step 2. Define permitted uses and scope

AI model training, search and recommendation, retrieval-augmented generation (RAG), and summarization services all use data differently.

Decide which uses you will allow, then specify the license period, covered services, whether redistribution is permitted, and retention and deletion terms. Pricing can be discussed together with these conditions.

Step 3. Prepare the data format and delivery method

Check whether content stored as documents or images can be delivered in a form buyers can use.

You may need text extraction and cleaning, deduplication, and metadata such as source and publication date. Also consider whether to deliver a one-time file, regular updates, or API access.

Step 4. Review demand and deal terms

Find out which companies need your data and what quality and delivery conditions they require.

Negotiate price and contract terms based on the rights you can grant and the data format you can provide, and settle how ongoing delivery and payment will work after the deal.

6. Check your data's deal potential with RDPGROUP

RDPGROUP helps organizations that own content and data assess its deal potential through RDPLINE and connect it with buyer demand.

The RDPGROUP program covers three stages:

Data value assessment: We review pricing rationale and expected prices under different deal conditions, based on volume, recency, scope of rights, supply stability and demand signals.
Sales terms: Based on the assessment, we define which data to supply and on what terms.
Buyer connection: We connect you with buyers who need your data and support the sale and delivery.

Use cases extend beyond AI model training to search, recommendation and data analytics. Which uses fit and what terms are possible depend on the characteristics and rights of your data.

Now is the time to set your terms

Global AI companies are already looking for Korean content. That gives every content owner a reason to review the rights and potential of what they hold.

Take stock of what you own, how far you can permit its use, and in what form you can supply it. With that groundwork, you will be ready to discuss concrete terms when demand comes knocking.

Curious about the deal potential and supply terms for your content and data? Find out with RDPGROUP.

Check your data's deal potential with RDPGROUP

References