Naver Cloud牵头财团获选韩国政府网络安全AI基础模型项目

Naver Cloud will develop a homegrown artificial intelligence model tailored to countering cyber threats against South Korea's critical national facilities.
The Naver Cloud consortium said on the 3rd that it has been selected as the operator for the cybersecurity segment of the government's specialized AI foundation model development project.
Naver Cloud and 32 other partners from industry, academia, research and the public sector plan to develop an AI model tailored to the domestic security environment and to verify and upgrade its performance by applying it at critical national facilities and major industrial sites.
Ahead of the formal launch of the project, the consortium has been pre-training the security-focused AI model using 4,000 of its own GPUs. In addition to the 256 GPUs provided by the government, the consortium will also draw on 4,000 Nvidia B200 chips from Naver Cloud and 256 H200 chips from LG. The consortium said this will shorten development time and allow more time for upgrading the model and for field testing at critical national facilities and industrial sites.
Using that infrastructure, the consortium will build a hyperscale, security-focused model with a 700-billion-parameter mixture-of-experts architecture, based on HyperCLOVA X and EXAONE. HyperCLOVA X will be developed as a separate model with strengthened defensive capabilities and EXAONE with strengthened offensive capabilities, giving the system full-cycle coverage from detecting cyberattacks to responding to them. The models' performance will be verified through two internationally recognized benchmarks.
For training data, which determines model performance, the consortium will use about 830 terabytes of high-quality real-world data that no single institution could assemble on its own. The pool spans operational data from Naver Cloud and LG CNS as well as core data from national infrastructure and major industries, including KEPCO KDN, Korea Hydro & Nuclear Power, the Financial Security Institute, KISTI and LG Uplus. Data refining technology covering threat intelligence collection, labeling, standardization and verification will be applied to train the models precisely on real threat environments.
The model will be continuously upgraded through testing at critical national facilities and major industrial sites. Field testing will be conducted in seven key areas — power, finance, science and technology, telecommunications, semiconductors, defense and aerospace — and the consortium aims to complete operating technology tailored to closed networks so the model can run reliably in security environments where external network access is restricted.
The consortium also plans to release the model developed through the project as open source available for commercial use, spreading the benefits across the domestic security industry. It will offer free vulnerability and cyberattack detection checkups for small and mid-sized companies while supporting domestic security firms in commercializing AI services through the Naver Cloud marketplace, building what it describes as a self-reinforcing national security AI ecosystem.
"Winning this project reflects comprehensive recognition of the large-scale AI infrastructure, data and security expertise the Naver Cloud consortium holds," Naver Cloud Chief Executive Kim You-won said. "We will take on this project with a sense of responsibility for building AI that protects the nation's critical infrastructure with our own technology, and we will create a national security AI model and ecosystem that represents South Korea."