Experts call on Korea to form 'DeepSeek-pursuing' task force
-
- PARK EUN-JEE
- [email protected]
-
- SEO JI-EUN
- [email protected]
-
- LEE JAE-LIM
- [email protected]
-
- SHIN HA-NEE
- [email protected]
![A monitor from an office computer at the governmental complex in central Seoul on Feb. 6 displays blocked access to China-based DeepSeek’s AI chatbot due to concerns of data privacy. [NEWS1]](https://koreajoongangdaily.joins.com/data/photo/2025/02/06/1a8711c8-87a6-4f76-bc5a-93e683522fd0.jpg)
A monitor from an office computer at the governmental complex in central Seoul on Feb. 6 displays blocked access to China-based DeepSeek’s AI chatbot due to concerns of data privacy. [NEWS1]
A group of Korean experts, engineers and professors urged the nation's government to form a centralized body in charge of supporting AI development, making a case that the advancement of Chinese startup DeepSeek could boost the morale of Korean firms with limited budgets and capabilities.
The demand followed a flurry of government agencies, tech and financial companies blocking access to DeepSeek's latest AI model on office computers due to security concerns.
Tech experts gathered at a seminar hosted by the Ministry of Science and ICT on Thursday to discuss the industry fallout and policy direction in light of the shocking rise of R1, DeepSeek's latest reasoning model.
“We need to form a pursuit team capable of advancing AI technology to the level of OpenAI and DeepSeek,” Prof. Kim Doo-hyun of Konkuk University’s Department of Computer Engineering and a member of the National Artificial Intelligence Committee said at a roundtable led by the committee and the Ministry of Science and ICT in central Seoul.
“A special task force should be established under the national AI computing center, free from regulatory constraints, to receive unprecedented support.”
Betting on AI computing
The ministry is already working to build a national AI computing center with an investment of 2 trillion won ($1.4 billion), aiming to support domestic companies and research institutes in their AI endeavors.
Forms of support include exempting selected companies from data-related copyright restrictions.
“These companies should be allowed to freely use domestic data for up to three years,” said Sung Kim, CEO of AI startup Upstage. “Copyright fees or other associated costs could be settled at a later stage.”
![A roundtable to collect insights on advancing domestic AI industry took place at the office of the National Artificial Intelligence Committee in central Seoul on Feb. 6 following the release of Chinese startup DeepSeek's latest reasoning model. Korean experts called for the establishment of a special task force to deliver “unprecedented” government support to designated AI companies. [NEWS1]](https://koreajoongangdaily.joins.com/data/photo/2025/02/06/df41eb7a-0a33-450e-a69f-6ef6404be24d.jpg)
A roundtable to collect insights on advancing domestic AI industry took place at the office of the National Artificial Intelligence Committee in central Seoul on Feb. 6 following the release of Chinese startup DeepSeek's latest reasoning model. Korean experts called for the establishment of a special task force to deliver “unprecedented” government support to designated AI companies. [NEWS1]
Another surprising revelation came from the head of LG AI Research, who claimed that domestic companies already possess the capability to train models at even lower costs than DeepSeek’s.
LG AI Research head Bae Kyung-hoon said it had cost 7 billion won to create the 32-billion-parameter Exaone 3.5, the latest large language model (LLM) released by LG’s research lab in December 2024. DeepSeek reported that $6 million was spent to train its V3 model.
“ChatExaone [an enterprise AI agent] operates similarly to DeepSeek’s Mixture of Experts system, a key factor that enabled the company to achieve low-cost development by combining multiple Exaone models,” he said. “The service is currently being used by all employees across LG affiliates. If it had been released globally beyond the group level and promoted more effectively, it could have gained greater recognition, which is somewhat regrettable.”
Bae claimed that per LLM, development costs less than 10 billion won, and the “entire development cost for one LLM can be summed up to less than 40 billion won.”
Bae promised to release an AI model that is as advanced as DeepSeek’s reasoning model R1 in open source “soon,” with hopes of government support that he expects to follow.
“OpenAI’s o3 surpasses R1 in inference capability, and reaching that level would require around 2,000 H200 GPUs and cost approximately 100 billion won,” he said. “Rather than distributing multitrillion won investments across multiple entities, the government should prioritize providing at least that level of funding to companies that are capable of making immediate progress.”
Tiptoeing to ban DeepSeek
Despite the positive feedback from the industry circles, many government agencies and public corporations are taking measures to prohibit people from accessing to the DeepSeek-made models via office computers.
The Ministry of Economy and Finance; Ministry of National Defense; Ministry of Trade, Industry and Energy; Ministry of Foreign Affairs; Ministry of Unification; Ministry of Health and Welfare; Ministry of Land, Infrastructure and Transport; and Ministry of Culture, Sports and Tourism either have limited or will limit access to the service offered by the Chinese startup, officials of the respective entities said in press briefings and phone calls.
The National Police Agency has also blocked access to DeepSeek's AI chatbot starting Thursday because of potential security risks.
![The logo of DeepSeek is displayed alongside its AI assistant app on a mobile phone, in this illustration picture taken Jan. 28, 2025. [REUTERS/YONHAP]](https://koreajoongangdaily.joins.com/data/photo/2025/02/06/57fb68cd-3462-4bf3-8a76-059975e3f03c.jpg)
The logo of DeepSeek is displayed alongside its AI assistant app on a mobile phone, in this illustration picture taken Jan. 28, 2025. [REUTERS/YONHAP]
The restriction follows a guideline issued Tuesday by the Ministry of the Interior and Safety on the use of generative AI. However, the guideline did not specify any particular services by name.
Korea Hydro & Nuclear Power (KHNP), the state-run operator of Korea’s nuclear and hydroelectric plants, uploaded a company-wide notice titled “Chinese AI service DeepSeek is banned” on Feb. 1.
KHNP had also prohibited employees from using ChatGPT for work-related tasks due to security issues surrounding the nuclear plants.
“The use of generative AI services for work tasks is banned to prevent leakage of company documents and personal information. We are not banning personal use, only use in a work environment,” KHNP said in the notice.
Privacy-sensitive
Financial authorities and institutions are also restricting employees’ access to DeepSeek’s AI services amid heightened security concerns over the Chinese startup’s privacy policy.
Following guidelines from the Interior Ministry, the Financial Services Commission and Financial Supervisory Service blocked access to DeepSeek’s services from their computers starting Thursday.
Banks, both private and state-run, also swiftly implemented blockades on their external networks over privacy concerns.
As for internal networks, any external websites are inaccessible within internal networks without prior approval as banks in Korea are required to abide by the network separation rules that mandate a physical separation of internal and external networks for security purposes.
KB Financial Group and KB Kookmin Bank blocked access starting Monday due to “concerns regarding its trustworthiness, given that there has been a database exposure at DeepSeek and that personal data collected by the firm is stored at servers in China and therefore would be subject to Chinese regulations,” said a spokesperson for the group.
Wiz, a New York-based cloud security company, said on Jan. 30 that it had identified a publicly accessible ClickHouse database from DeepSeek, “completely open and unauthenticated, exposing sensitive data.”
Hana Financial Group restricted access to DeepSeek on its external networks starting Tuesday, citing “concerns over excessive collection of data,” according to its spokesperson.
Woori Financial Group also restricted the chatbot on its external networks on Thursday.
Shinhan Financial Group requires prior approval for access of any external links even on their external networks, and therefore the access to DeepSeek is blocked by default, the firm explained. The Export-Import Bank of Korea blocked access to DeepSeek starting Jan. 31, and the Industrial Bank of Korea and Korea Development Bank followed suit.
Worries of 'excessive AI training'
Tech firms like Kakao, the operator of Korea’s dominant messenger app, are banning employees from using DeepSeek's R1 due to concerns that its latest open-source reasoning model is collecting “excessive amounts of information for AI training,” Yonhap reported Wednesday.
Kakao, which recently signed a partnership deal with OpenAI, sent out a company-wide notice to prohibiting “the use of DeepSeek for company-related tasks.”
The tech firm is the first major tech company to implement such a ban.
The decision was made due to security issues related to DeepSeek's's collection of users’ appliance information, intellectual property (IP) and keystroke dynamics, Kakao reasoned.
LG U+ also sent out a notice on Wednesday prohibiting employees from using DeepSeek's flagship model on the company's network due to security issues.
The notice also cautions workers against using DeepSeek on their personal computers until it is cleared to be safe for use. LG U+ also asked that employees refrain from putting important or sensitive information in queries.
The company’s server, does not allow access to the DeepSeek app, including for cloud tasks, as its domain has been suspended.
Naver is reportedly also considering forbidding employees from using DeepSeek for company tasks but has not issued a formal notice yet.
Ha Jung-woo, head of Naver’s Future AI Center, expressed his concerns about DeepSeek on social media in late January, after the Chinese startup released its latest large language model.
“Information about used equipment as well as keystroke patterns and rhythm, IP information and device IDs are saved to servers in China by default, not to mention the cookies,” Ha wrote.
Large firms like Samsung Electronics, SK Telecom and LG Electronics use proprietary generative AI models, and their company PCs block access for non-approved external programs.
Countries around the world are wary of DeepSeek, and their governments have been responding to the company's recent release on a state level.
Australia, Japan, Taiwan and the U.S. state of Texas have banned the use of the company's technology on government-owned devices while Italy has blocked it completely on app markets.
Britain and countries that are part of the European Union are also closely monitoring the China-developed technology.
Generative AI regulations are currently being fiercely debated in Korea. The Personal Information Protection Committee sent a formal request to DeepSeek to answer how, and in which categories, the company collects, stores and manages personal information.
Update, Feb. 6: Added information from a Ministry of Science and ICT roundtable.
Update, Feb. 6: Added that more ministries — such as the Ministry of Land, Infrastructure and Transport and the Ministry of Culture, Sports and Tourism — confirm the blocking of DeepSeek.
BY PARK EUN-JEE, LEE JAE-LIM, SEO JI-EUN, SHIN HA-NEE, JEONG JAE-HONG [[email protected]]
with the Korea JoongAng Daily
To write comments, please log in to one of the accounts.
Standards Board Policy (0/250자)