No AI training provisions in Daum's current terms of service
"User consent procedures are required for AI training use"

As Upstage acquires AXZ, which operates the portal site Daum, attention is focusing on whether the company will be able to use data to advance its artificial intelligence (AI) models without going through separate consent procedures for data use. This is because, in principle, the rights to data generated in services such as Daum Cafe belong to the users.


According to a review of Daum's terms of service by The Asia Business Daily on the 5th, the provisions dealing with the use of copyrighted works state that AXZ, which operates Daum, and its affiliates may use posts for the purposes of service improvement and research and development. However, there were no separate provisions established specifically for AI training. A representative from AXZ said, "That clause broadly covers general service enhancement, such as improving service quality or researching new services, and is not a provision related to AI training," adding, "If we are to use the data for AI training, separate terms of service may be necessary."


Participants are looking around the Upstage booth at the first presentation event of the "Proprietary AI Foundation Model" project held at COEX in Gangnam-gu, Seoul. Yonhap News

Participants are looking around the Upstage booth at the first presentation event of the "Proprietary AI Foundation Model" project held at COEX in Gangnam-gu, Seoul. Yonhap News

View original image

Kakao, the parent company of AXZ, and Upstage each held a board meeting on January 29 and approved the signing of a memorandum of understanding (MOU) for a share-swap transaction and related matters. Following due diligence by both sides, the core of the deal is that the AXZ shares held by Kakao will be transferred to Upstage, and Kakao will acquire a certain stake in Upstage. AXZ is currently a 100% subsidiary of Kakao.


The driving force behind Upstage's move to acquire Daum is the advancement of its AI models. The company plans to enhance its AI technologies, including its proprietary large language model (LLM) Solar, by leveraging Daum's diverse content data. Upstage has faced a shortage of Korean-language data in the process of AI model development and has so far overcome this by using a combination of various synthetic data and specialized data for training.


Launched in 1995 and regarded as a first-generation portal site, Daum holds a vast amount of content that users have generated over 30 years through services such as Daum Cafe and Tistory (blog). Although its presence in the search market is weak compared to Naver and Google, its community platforms where users generate content, such as Daum Cafe and Tistory, still retain a large number of users.


Daum launches dedicated tab for original short-form dramas. Provided by AXZ

Daum launches dedicated tab for original short-form dramas. Provided by AXZ

View original image

"In principle, permission from copyright holders is required for AI training"

Since Daum's current terms of service do not contain provisions related to AI training, a separate consent process is required in order to use the data for that purpose. Lee Sungyeob, a professor at Korea University's Graduate School of Management of Technology, said, "If you want to use data generated by users for AI training, you must obtain the users' consent," and explained, "In principle, you must obtain permission from the copyright holder who created the work for its use in AI training." An industry official also said, "As a rule, if you want to train an AI model on or otherwise utilize content produced by users, you need the users' consent."


Even in the case of news, one of Daum's main services, it is difficult to use the content immediately for AI model training. This is because the copyrights to news articles belong to each media outlet. In fact, the three terrestrial broadcasters have filed a lawsuit against Naver, claiming that the company used their articles without permission as data for AI training, and the trial is ongoing.


However, Upstage says that because the acquisition process is still underway, it is too early to comment specifically on AI use of the content. A representative from Upstage said, "We are currently about to begin due diligence for the acquisition, and since the acquisition has not yet been finalized, there are no concrete plans at this time," adding, "We will be able to share our plans once the full-scale acquisition process is completed."



Meanwhile, Upstage has previously expressed its intention to use data for AI model training in a legitimate manner. During a press conference last year, Upstage CEO Kim Sung-hoon said, "It is known that Chinese models and OpenAI did not pay any particular copyright fees for image training, and in that sense, I wonder if we are not competing in an unfair environment." He went on to say, "If discussions continue on ways to provide reasonable compensation for training data, we would like to participate," and stressed, "We prefer to do things legally, and we sign contracts with data providers and pay them a certain amount of money."


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