Ministry of Culture, Sports and Tourism and Korea Copyright Commission Release Fair Use Guidelines

Korea Copyright Commission Building

Korea Copyright Commission Building

View original image

It is a copyright infringement for artificial intelligence (AI) to learn from the full text of news articles and provide summary services. The same applies to training AI on textbooks to create new textbooks or workbooks. It is also illegal to purchase paid images, train AI on them, and generate new images, or to train AI on music tracks and create AI cover songs.


On December 4, the Ministry of Culture, Sports and Tourism and the Korea Copyright Commission held a public briefing at the National Folk Museum of Korea and announced the “Guide to Fair Use of Generative AI Training on Copyrighted Works.” As controversy arose over whether it constitutes fair use when AI learns from large volumes of copyrighted works to create new content, a special committee was formed in September to prepare this guide.


The guide outlines four conditions under which large-scale AI training does not qualify as fair use: infringement of the legitimate interests of the rights holder, lack of transformative use, absence of social or public interest, and commercial purpose.


In the case of news articles, training AI on the full text of articles without the consent of the media company and then providing commercial summary services does not constitute fair use. This is because it is difficult to recognize a transformative purpose and it causes economic harm to media companies.


It is also illegal to train AI on lawfully purchased textbooks to create new textbooks or workbooks. This is because there is no transformative purpose, and it may weaken the market dominance of publishers.


Training AI on paid images to generate new images also constitutes copyright infringement. This interferes with the technical protection measures of the producer and causes economic harm to the copyright holder.


Similarly, if AI is trained on songs purchased from music platforms to compose AI cover songs, there is a high likelihood that the resulting works will directly replace the original market and cause economic harm.


The guide also presents four cases that qualify as fair use: using public data as training material for natural language processing models; training AI on publicly available academic papers to provide summaries; training AI on tables and graph data from science and engineering papers that are accessible to anyone; and training AI on legally purchased videos for the purpose of analyzing criminal behavior patterns.



Kang Seokwon, Chairman of the Korea Copyright Commission, stated, "We have presented fair use interpretation standards that are suitable for the domestic situation," and added, "We hope this will help reduce some of the uncertainty related to AI training."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing