Naver Cloud to Build 700 Billion-Parameter Cybersecurity-Specialized AI (Full Story)
IT DAILY ·
✦ AI Summary
The Ministry of Science and ICT and the National IT Industry Promotion Agency said they had selected the Naver Cloud consortium as the operator of the project to develop a "cybersecurity-specialized artificial intelligence (AI) foundation model."
The competition was a two-way race between the Naver Cloud consortium and the SK Telecom consortium, and the Naver Cloud consortium was composed of 32 participating organizations in total.
The project includes support for 256 GPUs to develop a security-specialized AI foundation model, and Naver Cloud plans to release the model as open source and provide free inspection services and commercialization support.
The Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA) announced on the 3rd that they had selected the Naver Cloud consortium as the operator of the government's project to develop a "cybersecurity-specialized artificial intelligence (AI) foundation model."
The competition was a two-way race between the Naver Cloud consortium and the SK Telecom consortium. After the contest, the Naver Cloud consortium was confirmed as the final operator.
The project is a national initiative aimed at consortia made up of key domestic AI and cybersecurity companies. The support package includes 256 GPUs for developing a security-specialized AI foundation model.
The selected Naver Cloud consortium is composed of 32 participating organizations in total. The consortium will carry out the project titled "Development of an AI foundation model for internalizing national cyber defense capabilities."
The photo shows Naver Cloud CEO Kim Yoo-won. According to the caption, he is speaking at the NVIDIA Cloud Partner Summit.
LG initially pushed for an independent consortium, but LG CNS, LG AI Research, and LG Uplus ultimately joined the Naver Cloud camp, creating a large consortium that combines LLM research capabilities, large-scale SI expertise, and telecommunications infrastructure know-how.
The Naver Cloud consortium is already conducting pretraining for a security-specialized AI model before the official start of the project, using resources separate from the government-provided allocation. Naver Cloud has committed 4,000 B200s, and LG has made a preemptive investment by supplying 256 H200s. Through this, the consortium plans to shorten the development period, and the time saved will be devoted to improving model performance and field validation.
The key goal is to build a security-specialized model based on HyperCLOVA X and EXAONE with a final 700 billion-parameter (700B) MoE architecture. To this end, the company plans to strengthen defensive capabilities based on HyperCLOVA X and offensive capabilities based on EXAONE, and develop two independent models in parallel to secure full-cycle response capabilities for both attack and defense. In addition, it plans to rigorously verify objective performance using two internationally recognized benchmarks.
To secure model performance, it will use about 830 terabytes (TB) of high-quality real-world data for training. The data collected includes operational data from Naver Cloud and LG CNS, and the sources also include KEPCO KDN, Korea Hydro & Nuclear Power, the Financial Security Institute, KISTI, and LG Uplus. The data targets national infrastructure and key industrial assets, and the company plans to carry out precise learning in real threat environments by applying technologies for threat information collection, labeling, standardization, validation, and cleansing.
The developed model will be demonstrated in the field across the country's seven core strategic sectors. The demonstration areas are power, finance, science and technology, telecommunications, semiconductors, defense, and aerospace.
Naver Cloud plans to release the model developed under this project as open source for commercial use. It also plans to provide free vulnerability and cyberattack detection checks for small and midsize businesses, while supporting the commercialization of AI services from domestic security companies through the Naver Cloud marketplace.
LG CNS, a participating member of the consortium, said the project aims to secure a cybersecurity-specialized AI foundation model and, based on that, establish a foundation for developing a range of security services for domestic security companies and institutions, thereby contributing to the activation of the domestic AI security industry ecosystem.
After the project announcement, SK Telecom issued an official statement saying it had failed to win the contract, but described itself under its own rules as a full-stack AI company with AI infrastructure, models, and services. It added that it would continue efforts to strengthen South Korea's AI competitiveness and, regardless of whether it won government projects, would keep discussing continued cooperation with participating consortium members and steadily build its cybersecurity AI capabilities.
Source: IT DAILY · Seong Won-young
Original: https://www.itdaily.kr/news/articleView.html?idxno=241378
References
This article was produced with the help of an automated content generation algorithm.
Source: IT DAILY
View originalThis article was summarized and organized by BizCrush based on the original article from IT DAILY. For exact quotations and full details, please refer to the original article.