Developing a Korea-Specialized Security AI Model with 4,512 GPUs... Plans for Global Export (Comprehensive)

Ministry of Science and ICT Hosts Cybersecurity AI Partnership Meeting
Paik Kyunghoon: "Bold Action Needed, Not Slowing Down"
A Unified Team of 41 Organizations... Expanding Across Domestic and Global Industry Ecosystems

On the 22nd, at Lotte Hotel in Jung-gu, Seoul, Ministry of Science and ICT Vice Prime Minister and Minister Baek Kyung-hoon (fifth from the left in the front row) and other officials from consortium member organizations, including Kim Yoo-won, CEO of Naver Cloud, took a commemorative photo at the kickoff meeting for the development of a cybersecurity-specialized AI model project. 2026.9.22 Photo by Kim Hyun-min

Naver Cloud, which was selected for the government's "cybersecurity-specialized artificial intelligence (AI) model" project, announced plans to combine on-site data from domestic security companies to develop an AI cybersecurity model uniquely tailored to Korea's needs. The company aims to develop security AI models specialized for both offensive and defensive purposes using 4,512 graphics processing units (GPUs), with a focus on spreading the models domestically and aggressively exporting its sovereign AI security services abroad.

The Ministry of Science and ICT held a business meeting on the afternoon of September 22 at the Sapphire Ballroom on the 3rd floor of The Grand Lotte Seoul to hear Naver Cloud Consortium's development plans and discuss collaboration measures.

Naver Cloud, serving as the consortium leader, aims to develop a uniquely Korean, reliable AI cybersecurity model by combining security companies' real-world data and the expert knowledge of their defenders, creating a digital shield that can be immediately deployed at critical national facilities and industrial sites.

In his opening remarks, Hae-koon Bae, Deputy Prime Minister and Minister of Science and ICT, stated, "Securing our own AI security model as a matter of AI security sovereignty and as a strategic security asset is essential. Even as discussions emerge about controlling the pace of AI development, the possession of frontier AI capabilities will determine a nation's future, so we must continue with bold challenges rather than slowing down." Deputy Prime Minister Bae emphasized, "In this context, this project aims to swiftly and boldly develop globally competitive models deployable directly at actual security sites, enabling us to respond immediately to the cyber threats we face now."

Baek Ghun, Deputy Prime Minister and Minister of Science and ICT, speaks at the kickoff meeting for the Cybersecurity Specialized AI Model Development Project held on September 22, 2026, at Lotte Hotel in Jung-gu, Seoul. Photo by Kim Hyunmin

He continued, "We plan to work hand-in-hand with the cybersecurity-specialized model development consortium, supporting not just infrastructure, but also competitiveness, application, and proliferation, considering all aspects and providing active support. Through this project, we ask the public and private sectors to join forces so that Korea's AI can lead innovation on a foundation of safety and trust."

Kim Yuwon, CEO of Naver Cloud, said, "Most AI technologies currently used to respond to security threats are developed overseas. When data is not sufficiently provided, AI cannot function properly. That’s why we need a security-specialized model under our own control, and if we can create an environment where various local data and resources are used to train this model, we can ensure a much safer future."

CEO Kim: "Preparing for Widespread Use Beyond Participating Organizations"

CEO Kim added, "We are well aware that the outcome of this project should not be monopolized or used exclusively by Naver Cloud or participating organizations. We are preparing so that security companies, governmental agencies, schools, hospitals, and more in Korea can utilize it widely. After about ten months of preparation, we will strive to provide products of the highest quality."

Kim Yuwon, CEO of Naver Cloud, is speaking at the kickoff meeting for the Cybersecurity Specialized AI Model Development Project held at Lotte Hotel in Jung-gu, Seoul on September 22, 2026. Photo by Kim Hyunmin

Kim Jin-hwi, Executive Director at Naver Cloud, explained, "In the past, it could take 60 to 90 days from discovering a cybersecurity vulnerability to fixing it, but the rapid advancement of AI has reduced the average time to discover and resolve a vulnerability to about 45 minutes, which is approximately 2,000 times faster than before. In cybersecurity, AI is not an option but a necessity, and the nation must secure AI defense capabilities."

Naver Cloud has established three strategic pillars for the successful implementation of their proprietary cybersecurity-specialized AI foundation model: developing the technology in-house with proven methods over a ten-month period, validating its performance on the front lines of cybersecurity, and scaling it throughout the domestic and international industrial ecosystem.

Executive Director Kim stated, "A total of 41 institutions—including 33 consortium members and 8 cooperative organizations—have formed a unified team. Participating institutions are organized into division-specific teams for data collection and processing, model development, evaluation and validation, and field testing and demonstration. These teams will collaborate throughout the project period."

To secure competitiveness on both offense and defense, the plan is to simultaneously develop two specialized security AI models: ‘HyperCLOVA X’ by Naver Cloud, focusing on defensive capabilities, and ‘Exaone’ by LG AI Research, focused on offensive capabilities. For this, a total of 4,512 GPUs will be used—including 256 B200 GPUs provided by the government, and another 4,000 B200 GPUs in Naver Cloud's own cluster, deployed since August. LG AI Research also plans to deploy 256 H200 GPUs. The government-provided GPUs will be allocated according to each consortium member’s role—such as data preprocessing and model validation—and utilized as efficiently as possible.

Executive Director Kim further stated, "The Naver Cloud Consortium has collaborated to secure an unprecedented amount of cybersecurity data—about 830 terabytes—that would be impossible for a single organization to amass. We will use internationally recognized benchmarks to ensure the model’s performance can be measured quantitatively and objectively."

He added, "The specialized security models developed will be integrated with various solutions and APIs from Korean security companies. We will develop services for AI-based vulnerability detection, AI-driven cyberattack response, and AI-powered incident investigation, interlinking them with domestic security products." The two resulting specialized models will be released as open source, and vulnerability assessment services will be offered to small- and medium-sized businesses lacking sufficient security staff or budgets. After validation, efforts will focus on expanding these models across industrial ecosystems both domestically and internationally, with ambitions to export sovereign AI security services from the internet backbone to key national infrastructure on a global scale.

Executive Director Kim emphasized, "We will defend Korea with Korean security-specialized AI models that overseas models cannot provide. The Naver Cloud Consortium is determined to successfully develop a proprietary AI security foundation model to internalize national cybersecurity capabilities."

This content was produced with the assistance of AI translation services.