On September 3, the Ministry of Science and ICT selected the Naver Cloud consortium as the organization to carry out a project to develop a cybersecurity-specialized AI foundation model.
The goal of the project is to develop a domestic security model capable of responding to hacking threats that use AI.
However, the selection announcement is a decision to begin a 10-month project, not an announcement of development results or a public service launch.
3-line summary
1. Naver Cloud was selected as the operator of a security AI project
2. It will receive support for 256 GPUs for 5 months at a time
3. Whether support will be extended will be decided after the interim evaluation
Developing Models for Both Defense and Attack Analysis
The Ministry of Science and ICT announced that it had selected the Naver Cloud consortium as the final organization to carry out the “Development of a Cybersecurity-Specialized AI Foundation Model” project. The government stated that its objective was to secure a domestic security model capable of responding to hacking threats that use AI.
The consortium plans to develop 2 models in parallel, based on Naver Cloud’s HyperCLOVA X and LG AI Research’s EXAONE, with enhanced defense and attack capabilities. Here, attack capabilities refers to the goal of developing models that analyze security vulnerabilities and attack methods; no completed functions or performance results have actually been announced.
Participating organizations include LG CNS, LG Uplus, the Financial Security Institute, KEPCO KDN, Korea Hydro & Nuclear Power, and various security and AI companies and research institutions. The government explained that it positively evaluated the development strategy of linking and integrating multiple AI agents and the industry-academia-research cooperation system.
The Support Period Is 10 Months, but There Is an Evaluation After Halfway
The selected consortium will receive 256 B200 high-performance graphics processing units from NVIDIA. The period is 10 months starting this month, but support for the entire period was not confirmed from the outset.
The government plans to provide GPUs for the first 5 months, conduct an interim evaluation, and decide whether to provide support for the remaining 5 months based on the evaluation results. Therefore, what has been confirmed in this announcement is the selection of the project organization and initial support; it is difficult to regard the release date, performance, or commercialization scope of the final model as having been confirmed.
Naver Cloud stated that it had secured 830 terabytes of training data and had been conducting pretraining using 4,000 (4000) B200 units it owns and 256 H200 units provided by LG. This is the consortium’s explanation and is not yet at a stage where the figures can be interpreted as guaranteeing the development results.
“Open-Source Release” Has Been Presented as a Project Goal
The consortium announced plans to release the completed model in an open-source format that can be used commercially, as well as to provide small and medium-sized companies with free inspection services for detecting vulnerabilities and cyberattacks. It also presented a marketplace concept to support the commercialization of AI services by domestic security companies.
The point of interest in this project is that it seeks to create a separate model tailored to security work, rather than a large-scale general-purpose AI. How well a system that connects AIs with different roles—such as threat detection, vulnerability analysis, attack analysis, and response—will function in actual security settings will have to be confirmed through the development and evaluation processes.
References
Tags #AI #ArtificialIntelligence #SecurityAI #Cybersecurity #NaverCloud #HyperCLOVAX #EXAONE #MinistryOfScienceAndICT #GPU #NVIDIAB200 #AIFoundationModel #VulnerabilityAnalysis