South Korean cybersecurity AI development project

The Naver Cloud consortium has been selected by South Korea’s Ministry of Science and ICT to develop a specialized AI foundation model for cybersecurity. The project aims to internalize national cyber security capabilities by building massive AI models with approximately 700 billion parameters using a Mixture-of-Experts (MoE) architecture.

The development strategy utilizes a dual-track approach: Naver Cloud’s ‘HyperCLOVA X’ will focus on defensive capabilities, while LG AI Research’s ‘EXAONE’ will focus on offensive capabilities. This parallel development is intended to better understand and counter attacker movements. The consortium, which beat out a competing bid from SK Telecom, includes 32 to 33 institutions, such as LG CNS, LG Uplus, and various academic and research organizations.

To support the project, the government will provide 256 B200 GPUs. The consortium is also utilizing significant private computing power, including 4,000 NVIDIA B200 GPUs from Naver Cloud and 256 H200 GPUs from LG. The models will be trained on approximately 830 terabytes of real-world data from sectors such as power, finance, telecommunications, and defense.

Once developed, the AI will be tested in seven critical sectors, including defense, aerospace, and semiconductors, with a focus on operating within closed networks to ensure data security. Additionally, the consortium plans to release the models in an open-source format for commercial use to foster a domestic cybersecurity AI ecosystem.