Project: Cybersecurity-Specialized AI Foundation Model Development
Organizer: Ministry of Science and ICT (MSIT)
Executing Organization: NAVER Cloud Consortium (32 participating institutions)
Period: September 2026 – February 2027 (Phase 1)
Original News1 article (Korean)
The development of a domestic AI foundation model specialized for cybersecurity is now under way in earnest. The goal is to secure, through domestic technology, a security AI capable of protecting national critical infrastructure without relying on general-purpose foreign models. Our lab is among the participating institutions in the consortium.
The Ministry of Science and ICT announced that it has selected the NAVER Cloud consortium as the final executing organization for the ‘Cybersecurity-Specialized AI Foundation Model Development’ project. The decision followed presentation evaluations of the two bidders that entered the open call.
A total of 32 institutions participate in the consortium. LG ICT affiliates including LG CNS, LG AI Research, and LG Uplus have joined, along with leading domestic security companies such as SANDS Lab, S2W, Logpresso, ESTsecurity, and Sparrow. Major industrial and public institutions also take part, including Korea Aerospace Industries (KAI), Korea Hydro & Nuclear Power, KEPCO KDN, the Financial Security Institute, and the Korea Institute of Science and Technology Information (KISTI). From academia, Chung-Ang University participates alongside Seoul National University, KAIST, Korea University, POSTECH, Korea Institute of Energy Technology, and Daegu University.
The target is a hyperscale security-specialized model with a 700B-class Mixture-of-Experts (MoE) architecture. The plan is to develop two independent models in parallel, one strengthening defensive capabilities and the other offensive capabilities, so as to secure coverage across the full attack-and-defense cycle. Training draws on roughly 830TB of high-quality real-world data, processed through threat intelligence collection, labeling, standardization, and validation so that the model learns actual threat environments in detail.
The resulting model will be validated across seven domains: power, finance, science and technology, telecommunications, semiconductors, defense, and aerospace. The consortium stated that it will release the completed model as open source available for commercial use, and will also provide free security assessment services for small and medium-sized enterprises.
The government will provide 256 NVIDIA B200 GPUs over the 10-month project period, worth approximately 16.9 billion KRW. Five months of support are provided first, with the remaining five months determined by the results of an interim evaluation. Phase 1 runs from September 2026 to February 2027, with interim results disclosed and reviewed in February 2027 and the final report due in July 2027.
Yuwon Kim, CEO of NAVER Cloud, said, “We will approach this project with a sense of responsibility for building, with our own technology, an AI that protects national critical infrastructure, and we will create a national security AI model and ecosystem that represents Korea.”