What is the fair price for a book used in artificial intelligence (AI) training? As the generative AI industry rapidly expands, debates over appropriate usage fees for publishing data are intensifying. In the U.S., copyright disputes have led to settlements worth millions of dollars per book, and similar pricing issues are emerging in South Korea.
According to the publishing industry, the Korean Publishers Association (KPA) plans to distribute a revised standard contract to publishers this month, which will include provisions regarding the use of data for AI training. This revision adds AI-related clauses to existing contracts between authors and publishers, and it is currently undergoing legal review. The aim is to clarify the scope of usage and rights concerning works used for AI training to prevent future disputes.
A KPA official stated, "We are conducting a legal review of the amendments to the standard contract that will include clauses related to publishing rights and AI training data. Once the review is complete, we plan to provide relevant guidelines to publishers as early as this month."
This initiative comes amid growing transactions involving publishing data for AI training, which have sparked controversies over copyright and appropriate usage fees.
Recently, the KPA asserted that the settlement amount reached with Woongjin Booksen in 2023 regarding the National Institute of the Korean Language's corpus project should not serve as a general pricing standard for AI training data. The settlement was a unique agreement considering the public interest of the project, the resolution of copyright disputes, and the usage period until 2027, making it unsuitable for commercial AI training data transactions.
The controversy stems from an incident involving the unauthorized use of 15,933 published works in the National Institute of the Korean Language's corpus construction project. Although the KPA reached a compensation agreement with Woongjin Booksen, an appeals court later considered this settlement amount as a key basis for determining damages, resulting in a lower compensation amount than the initial ruling. The case is currently pending before the Supreme Court. The KPA maintains that this agreement reflects the unique circumstances of a government project focused on public interest and dispute resolution, and interpreting it as a pricing standard for AI training data is a misjudgment.
In South Korea, transactions involving publishing data for AI training are ongoing. The KPA has supplied publishing data for government-supported projects related to independent AI foundation models and the medical sector. Companies like Naver and SK Telecom have also entered contracts with individual publishers. Specific contract terms and transaction amounts have not been disclosed, but they are reportedly higher than the previous settlement amount with Woongjin Booksen.
Internationally, there have been instances of settlements and contracts for AI training books reaching millions of dollars per title. The U.S. AI company Anthropic established a $1.5 billion settlement in a copyright class action lawsuit over illegal book downloads, agreeing to pay at least $3,000 per title for approximately 500,000 books. HarperCollins, a U.S. publisher, reportedly proposed a contract with Microsoft in 2024 for the AI training use of certain nonfiction books, offering to split $5,000 per title between authors and publishers. However, it is important to note that Anthropic's settlement is related to copyright infringement, while HarperCollins' agreement is based on prior usage permissions.
The KPA is also considering legal action regarding Anthropic's acquisition of Korean books. It has been determined that an intermediary distributor secured around 300,000 Korean books from Woongjin Booksen and domestic secondhand bookstores for export to the U.S. The publishing industry believes these books are likely to have been used for AI training. However, applying domestic law to actions conducted abroad presents challenges. The KPA is legally reviewing the possibility of domestic litigation while exploring collaborative responses with organizations like the Korean Publishers Conference.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.