The Naver Cloud consortium has been selected as the final performing organization for the government's security-specialized artificial intelligence (AI) project. The plan is to gather domestic AI infrastructure and security data to develop security AI models specialized in attack and defense, respectively, and apply them to critical national infrastructure and major industrial sites.
The Ministry of Science and ICT announced on September 3 that it has selected Naver Cloud as the contractor to execute the “Development of Cybersecurity-Specialized Artificial Intelligence (AI) Foundation Models.”
The government highly evaluated the fact that Naver Cloud formed a large-scale industry-academia-research consortium with outstanding technical capabilities and development experience, and presented a development strategy utilizing multiple agents and harnesses.
The consortium consists of a total of 33 organizations, including Naver Cloud, LG CNS, LG AI Research, LG Uplus, and major AI and security companies. In addition to security-related public and research institutions and universities, actual demand organizations such as Korea Aerospace Industries, Ltd. (KAI), Korea Hydro & Nuclear Power Co., Ltd. (KHNP), and KDN Co., Ltd. (KDN) also participated.
While SK Telecom Co., Ltd. strengthened model competitiveness by partnering with Upstage in the independent AI foundation model project, Naver Cloud chose a strategy to gather not only models but also infrastructure, data, and demand organizations into a single consortium.
This structure, which allows learning from actual security field data and immediate verification of developed models, is analyzed to have served as a strength in terms of project execution capability.
The computing resources being committed are also large-scale. Naver Cloud proactively deployed 4,000 unit of its own Nvidia B200 graphics processing units (GPUs) prior to the official project launch to conduct pre-training. It plans to additionally deploy 256 LG H200 units separately from the 256 government-supported B200 units.
Based on this, it will develop two security-specialized models with a 700B-class (700 billion parameters) Mixture-of-Experts (MoE) structure utilizing “HyperCLOVA X” and “EXAONE,” respectively. It plans to use approximately 830 terabytes (TB) of actual security-related data for learning and conduct field demonstrations across seven sectors: power, finance, science and technology, telecommunications, semiconductors, defense industry, and aerospace.
Considering the security environment of critical national infrastructure where external network access is restricted, the model will be sophisticated to operate stably even in closed networks.
The models that have undergone field demonstrations will be released as open source for commercial use, aiming to contribute to the commercialization of AI services by domestic security companies.
The government will support Naver Cloud with 256 B200 GPUs for 10 months starting this month. After an interim evaluation in five months, a decision will be made on whether to support the remaining five months.
Kim Yu-won, CEO of Naver Cloud, said, “Winning this project order is the result of comprehensive recognition of the large-scale AI infrastructure, data, and security expertise possessed by the consortium,” adding, “With a sense of responsibility to safeguard critical national infrastructure using our own technology, we will build a national security AI model and ecosystem representing the Republic of Korea.”
