NAVER Cloud Begins Developing Two 700B-Class Cybersecurity-Specialized AI Models

[Financial News] The NAVER Cloud Consortium has begun full-scale development after being selected as the final operator of the government's project to develop a specialized artificial intelligence (AI) foundation model in the field of cybersecurity. The consortium is building AI models optimized for South Korea's cybersecurity environment. Its plan is to validate and enhance the models at national core facilities and major industrial sites, while releasing them as open source to raise the competitiveness of South Korea's cybersecurity AI industry.
According to industry sources on the 3rd, the Ministry of Science and ICT selected the NAVER Cloud Consortium as the final operator in a competition to develop a cybersecurity-specialized AI foundation model.
The NAVER Cloud Consortium began pretraining its cybersecurity-specialized AI models by deploying 4,000 graphics processing units (GPUs) ahead of the official launch of the project. In addition to the 256 GPUs provided by the government, the consortium is deploying 4,000 B200 GPUs from NAVER Cloud and 256 H200 GPUs from LG.
The move is intended to significantly shorten the development period through aggressive early investment. The time saved can be devoted to improving model performance and conducting field demonstrations, thereby maximizing the models' completeness.
The NAVER Cloud Consortium will build a massive cybersecurity-specialized model with a mixture-of-experts (MoE) architecture and approximately 700 billion parameters (700B), based on HyperCLOVA X and EXAONE. It will develop two proprietary models in parallel, strengthening defensive capabilities through HyperCLOVA X and offensive capabilities through EXAONE. The consortium plans to secure full-cycle capabilities for responding to cyberattacks and defenses, and to rigorously verify objective performance through two internationally recognized benchmarks.
The training data that will determine the models' performance consists of approximately 830 terabytes (TB) of high-quality real-world data, a volume difficult for a single organization to secure. It encompasses operational data from NAVER Cloud and LG CNS, as well as key data from national infrastructure and major industries held by KEPCO Knowledge, Data & Network (KEPCO KDN), Korea Hydro & Nuclear Power (KHNP), the Financial Security Institute, the Korea Institute of Science and Technology Information (KISTI), and LG Uplus. The consortium will apply data-refinement technologies ranging from threat-information collection to labeling, standardization, and verification to train the models precisely in real-world threat environments.
The models will be continuously enhanced through field demonstrations at national core facilities and major industrial sites. Demonstrations will be conducted across seven major sectors—power, finance, science and technology, telecommunications, semiconductors, defense, and aerospace. The consortium also plans to complete operational technologies specialized for closed-network environments, enabling stable operation even where access to external networks is restricted.
The NAVER Cloud Consortium will release the models developed through this project as open source for commercial use. It plans to spread the results throughout South Korea's cybersecurity ecosystem. In addition to providing small and midsize companies with free vulnerability and cyberattack detection assessments, the consortium will actively support the commercialization of AI services by South Korean cybersecurity companies through the NAVER Cloud Marketplace.
Yuwon Kim, CEO of NAVER Cloud, said, "Winning this project is the result of comprehensive recognition of the NAVER Cloud Consortium's large-scale AI infrastructure and data, as well as its cybersecurity expertise." Kim added, "We will approach the project with a sense of responsibility for creating, with our own technology, AI that protects the nation's core infrastructure, and build a national cybersecurity AI model and ecosystem representing South Korea."
[email protected] Jang Min-kwon Reporter