韩国科技部选定Naver Cloud联盟主导网络安全专用AI基础模型开发项目，击败SKT联盟

Naver Cloud has beaten SK Telecom to lead a government project to develop an artificial intelligence (AI) foundation model specialized for cybersecurity.
The two companies also competed in the government’s Sovereign AI Foundation Model project, but with contrasting results. SKT passed both the first and second stage evaluations, while Naver Cloud was eliminated in the first stage evaluation in January. In the latest security-focused AI project, Naver Cloud was selected as the final contractor over SKT.
The Ministry of Science and ICT (MSIT) said Sept. 3 that it selected the Naver Cloud consortium as the final organization to carry out the “Cybersecurity-Specialized AI Foundation Model Development” project.
The evaluation involved external experts in AI and cybersecurity. Based on each consortium’s business proposal and presentation, the evaluators assessed technology capabilities and development experience, development strategies and technologies, the quality of development goals, feasibility, market potential and broader impact. MSIT said the Naver Cloud consortium received favorable evaluations for establishing a large-scale industry-academia-research collaboration framework and for its development strategy of connecting and integrating multiple AI agents.
The rival SKT consortium was also a strong contender based on its composition. Its AI model development team was particularly notable. SKT and Upstage were among the teams that passed the second stage evaluation of the government’s Sovereign AI Foundation Model project. Two teams that passed that evaluation joined forces in a single consortium for the cybersecurity AI project. The consortium also included nine cybersecurity companies: SK Shieldus, SECUI, AhnLab, IGLOO, RaonSecure, SmartM2M, AIM Intelligence, Genians and PIOLINK. Korea University, Soongsil University’s Industry-Academic Cooperation Foundation and the Korea Information Security Industry Association (KISIA) also participated, bringing the consortium to 14 organizations including SKT.
The Naver Cloud consortium was similarly strong in cybersecurity expertise. Led by Naver Cloud, it includes LG CNS, LG AI Research and LG Uplus, as well as cybersecurity and AI companies including S2W, ESTsecurity and JiranSecurity. The Financial Security Institute, KEPCO KDN, Korea Hydro & Nuclear Power, the Korea Institute of Science and Technology Information (KISTI), universities and research institutions are also participating.
Naver Cloud proposed a specific model-development approach. The consortium will develop two models in parallel, specializing respectively in defensive and offensive capabilities, based on Naver’s HyperCLOVA X and LG AI Research’s EXAONE. The ultimate goal is to build a cybersecurity-specialized model with a 700-billion-parameter mixture-of-experts (MoE) architecture.
The government highlighted the use of “multiple agents and a harness” in announcing the selection. The agents refer to AI systems responsible for different cybersecurity tasks, including threat detection, vulnerability analysis, attack analysis and response. The harness is an integrated execution framework that connects and manages the operation of those agents, AI models and external security tools. The government said it viewed the Naver Cloud consortium’s development strategy and objectives favorably.
MSIT, however, did not disclose scores for individual evaluation categories or the difference between the competing consortiums’ scores, making it unclear which specific factors ultimately determined the outcome.
SKT plans to continue developing cybersecurity AI separately from the project. The company said it will discuss future cooperation with the organizations that participated in its consortium and continue building cybersecurity AI capabilities regardless of the government project.
The selected Naver Cloud consortium will receive access to 256 Nvidia B200 GPUs for 10 months starting this month. The GPUs will initially be provided for five months, with support for the remaining five months to be determined based on the results of an interim evaluation.