Naver Cloud to Develop AI Specialized in Cybersecurity, Backed by 33-Entity Industry-Academia Consortium
TECHWORLD ·
✦ AI Summary
The Ministry of Science and ICT announced that it had selected the Naver Cloud consortium as the operator of the "Development of an AI Foundation Model Specialized in Cybersecurity" project.
The consortium consists of 33 organizations, with Naver Cloud as the lead organization and participants including LG CNS, LG AI Research, LG Uplus, the Korea Financial Security Institute, Korea Hydro & Nuclear Power, KDN, KISTI, and KAI.
Starting this month, the government will provide 256 NVIDIA B200 GPUs for 10 months, while the consortium will use its own GPUs and data to develop a cybersecurity-specialized AI model and carry out field demonstrations.
The Ministry of Science and ICT announced on the 3rd that it had selected a Naver Cloud consortium as the operator of the "Development of an AI Foundation Model Specialized in Cybersecurity" project. Two consortia participated in the public call, and the ministry evaluated technology, development experience, development strategy, feasibility of achieving goals, marketability, and ripple effects with input from external experts in AI and cybersecurity. Naver Cloud's consortium was ultimately selected.
The Naver Cloud consortium is made up of 33 organizations in total. Naver Cloud is the lead organization. Participating organizations include LG CNS, LG AI Research, LG Uplus, the Korea Financial Security Institute, Korea Hydro & Nuclear Power, KDN, the Korea Institute of Science and Technology Information (KISTI), and Korea Aerospace Industries (KAI).
Naver Cloud will team up with 32 organizations, including LG CNS and LG AI Research, to begin developing an AI foundation model specialized in South Korea's security environment. The consortium consists of security companies, AI companies, operators of national infrastructure, universities, and research institutions.
The consortium will move forward from developing an AI foundation model tailored to the domestic security environment to field demonstrations, with a wide range of participating organizations working together.
Starting this month, the government will provide the Naver Cloud consortium with 256 NVIDIA B200 GPUs for 10 months, with the initial support period set at 5 months. A further 5 months of support will be decided based on the results of an interim evaluation.
In addition to government support, the consortium will deploy its own AI infrastructure. Before the project even begins, it will conduct pretraining for a security-specialized model using 4,000 B200 GPUs owned by Naver Cloud, and it plans to additionally use 256 H200 GPUs from LG. The computing resources secured in this way will be used for model training, performance improvements, and field demonstrations.
The base models for development are Naver's "HyperCLOVA X" and LG AI Research's "EXAONE," and the final application architecture will be a 700B-class Mixture of Experts (MoE). Two models will be developed in parallel: the HyperCLOVA X-based model will focus on defensive capabilities, while the EXAONE-based model will strengthen security analysis from an attacker's perspective.
Security data for training a large-scale model will be jointly built. The sources of training data will be operational data from Naver Cloud and LG CNS, along with data held by KDN, Korea Hydro & Nuclear Power, the Korea Financial Security Institute, KISTI, and LG Uplus, and the amount of training data secured is about 830TB. The consortium said it will collect threat intelligence, then go through data labeling, data standardization, and data verification, and use the verified data for model training.
The developed model will be demonstrated in 7 sectors: power, finance, science and technology, telecommunications, semiconductors, defense, and aerospace. It will take into account environments with restricted internet connectivity, and the target settings are national critical facilities and major industrial sites. The consortium will also develop technologies that can operate in closed networks.
After field validation, the consortium will work to expand the scope of use across South Korea's security industry. The consortium's developed model will be released as open source, and commercial use of the released model will be allowed. It also laid out plans to support vulnerability checks for small and medium-sized enterprises and to support cyberattack detection checks for small and medium-sized enterprises, and said it will also support the commercialization of AI services by domestic security companies through the Naver Cloud Marketplace.
The ministry plans to sign an agreement with the consortium within this month and then establish a working-level consultative body. After forming the body, the ministry plans to begin model development and also start GPU support.
Naver Cloud also outlined its direction for participating in the project. Kim Yoo-won, CEO of Naver Cloud, expressed his determination to build AI to protect the country's core infrastructure with homegrown technology, and described the goal as creating a national security AI model and ecosystem.
Source: TECHWORLD · Kim Seung-ki
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406475
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.