Naver Cloud Consortium Launches Development of AI Models Specialized in Domestic Security Environment
Naver Cloud Consortium has announced that it has been selected as the final business operator for the government's "Development of Specialized Artificial Intelligence Foundation Models (Cybersecurity Sector)" project and is commencing full-scale development.
Through this project, the consortium, comprising a total of 33 companies, academic institutions, research organizations, and public agencies, plans to construct an AI model optimized for Korea's security environment and advance domestic security AI competitiveness by conducting verification and enhancement at national critical facilities and major industrial sites.
Notably, the consortium has been conducting pre-training of security-specialized AI models by proactively deploying 4,000 GPUs it possesses even before the official project launch. In addition to the government-provided capacity of 256 units, the structure involves deploying Naver Cloud's 4,000 B200 units and LG's 256 H200 units. Building on this comprehensive advance investment, the development period will be significantly shortened, with the secured time concentrated on model performance enhancement and field demonstration to maximize completeness.
Based on this large-scale infrastructure, the consortium will construct an ultra-large-scale security-specialized model with a final 700B-level MoE (Mixture of Experts) structure based on HyperClovaX and Exaone. By developing two proprietary models in parallel—one reinforcing defensive capabilities (HyperClovaX) and the other attack capabilities (Exaone)—the consortium will secure comprehensive response capabilities across the entire attack and defense cycle, with plans to rigorously verify objective performance using two internationally recognized benchmarks.
Additionally, the learning data that determines model performance will utilize approximately 830TB of high-quality real data, which is difficult for a single institution to secure. This encompasses operational data from Naver Cloud and LG CNS, as well as core data from national infrastructure and major industries including Korea Electric Power Corporation KDN, Korea Hydro and Nuclear Power, Financial Security Institute, KISTI, and LG U+ . Data refinement technology spanning from threat intelligence collection to labeling, standardization, and validation will be applied to enable precise learning of actual threat environments.
The model will be continuously enhanced through demonstration at national critical facilities and major industrial sites. Field demonstrations will be conducted in seven major sectors—power, finance, science and technology, telecommunications, semiconductors, defense industry, and aerospace—and specialized operating technology for "closed networks" will be completed to ensure stable operation even in security environments with restricted external network access.
Kim Yoo-won, CEO of Naver Cloud, stated, "This project win is recognition of the Naver Cloud Consortium's comprehensive AI infrastructure, data, and security expertise. We will undertake this project with a sense of responsibility in creating AI that protects Korea's national critical infrastructure, and will develop a national security AI model and ecosystem representative of the Republic of Korea."