Naver Cloud has won the government contract to build a cybersecurity-specialized AI foundation model, beating out a competing bid from SK Telecom. The project addresses the rising threat of AI-powered cyberattacks by creating a domestically developed model for threat analysis, vulnerability detection, and incident response. The Ministry of Science and ICT announced the selection on September 3, with development set to begin this month.
The consortium and its approach
Naver Cloud's bid received high marks for its development strategy, which includes a large-scale industry-academia-research collaboration and the use of multiple AI agents and harnesses - software systems that connect AI models with external tools, plan task sequences, and coordinate functions. The consortium brings together 32 institutions, including security and AI firms LG CNS, LG Uplus, Sands Lab, and ESTsecurity, alongside universities such as KAIST, Seoul National University, Korea University, and POSTECH.
An external expert committee evaluated proposals based on technical capabilities, development strategy, feasibility of goals, and marketability. The government will provide 256 NVIDIA B200 GPUs across 32 nodes for 10 months starting next month, though support for the second five months depends on an interim evaluation. A ministry official said, "We plan to form an immediate working-level consultative body with the selected business operator to strive for swift results."
What the model will do
Naver Cloud's consortium said it will develop an AI model tailored to the domestic security environment and run demonstrations at national core facilities and major industrial sites. Built on HyperCLOVA X and Exaone, the project aims to produce two cybersecurity AI models with a 700-billion-parameter mixture-of-experts architecture. Beyond the government-supplied GPUs, Naver Cloud will deploy 4,000 B200 GPUs and LG will contribute 256 H200 GPUs. Training will draw on roughly 830 terabytes of data from Naver Cloud, LG CNS, and national infrastructure and industrial institutions.
The bigger picture
The push for a homegrown cybersecurity AI model reflects growing concern that offensive AI tools are becoming more sophisticated and accessible. By building specialized models rather than adapting general-purpose systems, security teams can train on threat data specific to national infrastructure and industrial networks. The project also signals South Korea's intent to reduce reliance on foreign-built AI for critical defense functions. For professionals working in AI for IT & Development, the move underscores how domain-specific models are reshaping security operations.
Why this matters for IT and development professionals
This project will produce models designed for vulnerability detection and incident response - tasks that directly affect security analysts, SOC engineers, and developers building secure systems. The 700B-parameter MoE architecture and multi-agent coordination approach point to a future where AI doesn't just flag anomalies but orchestrates response workflows. For teams evaluating AI-driven security tools, the techniques tested here - from training data sourcing to agent-based task planning - will influence what commercially available products look like within two years. Those interested in the intersection of machine learning and threat detection can explore AI for Cybersecurity Analysts to build relevant skills now.
Your membership also unlocks:
