
The government is collaborating with 41 organizations to develop a national security AI model specialized for attack and defense, and is piloting an AI grocery shopping service called 'Effective Shopping Map.'
AI-generated summary
With the rapid development of AI technology and growing concerns about abuse, the need to develop a security AI model at the national level is emerging.
41 organizations participate to promote attack/defense specialized model
First AI grocery shopping service... Expanding usability in daily life
(Seoul = Yonhap News) Reporter Park Sang-hyun = As the possibility of artificial intelligence (AI) escaping human control and being abused for crimes increases, the importance of AI security sovereignty is increasing.
As it is certain that the level of cyber threats will increase, our government has also begun developing AI models specialized in security.
The plan is to create a world-class 'cyber shield' that can be used immediately in security settings without being swayed by the idea of controlling the speed of AI development raised by some.
◇ Security AI model emerged as a security strategic asset
The domestic cyber security AI model is developed by a consortium led by Naver Cloud.
Naver Cloud plans to simultaneously create two security AI models specialized in attack (offensive) and defense (defensive) capabilities.
The world's highest level frontier model was set as the benchmark goal for each model.
LG AI Research Institute's ExaOne, which strengthens its attack capabilities, aims to reach 100% of China's Zhifu 'GLM 5.3' model in 'CyberZym', the highest benchmark in the field of vulnerability discovery, by October 2027.
Naver Cloud's Hyperclova
This project is promoted by a total of 41 organizations, including 33 participating organizations and 8 partner organizations, forming a team.
The government plans to complete a security-specific AI model within 10 months and distribute it to the government, businesses, schools, hospitals, etc.
Deputy Prime Minister and Minister of Science and ICT Bae Baek-hoon said at a meeting on the 22nd, "It is essential to secure our own security AI model as a security strategic asset," and "Through this project, we will lead AI innovation based on the foundation of safety and trust."
◇ Independent security AI is essential due to unique environmental and geopolitical characteristics
The Korea Internet & Security Agency (KISA) also expressed the opinion that we must hurry to build an AI-based defense system and develop a Korean-style security-specific foundation model.
KISA is conducting vulnerability discovery and analysis using AI on public web pages of major infrastructure, commercial software, and open source through the Open AI Cooperation Program.
We plan to significantly shorten the work period, which previously took several weeks, using AI, and support complementary measures for vulnerabilities identified through analysis.
Bae Seung-kwon, head of KISA's Vulnerability Management Center, said on the 21st of this month that the number of vulnerability disclosures is rapidly increasing due to the use of AI, but it is important to select high-risk issues and respond quickly rather than responding equally to all vulnerabilities.
He said, “The paradigm of vulnerability management is moving from simple discovery to verification and rapid action,” and suggested that companies should also pay attention to software supply chain management and application of the principle of least privilege when introducing AI.
Kim Eun-seong, head of the KISA Threat Response Policy Team, pointed out the need for an independent AI model specialized in the security field.
He said, “Korea has a unique security environment such as its own network partition system, document format (HWP), and document encryption (DRM), and it also has great geopolitical specificity,” and emphasized that a secure AI model is essential to build a system to early identify and delay AI model attacks.
◇ Food ingredient purchasing information at a glance… ‘I might shop frugally’
In the case of the 'AI Top 10 Projects for People's Livelihood', which the government is promoting to increase the use of AI in daily life, results are gradually being produced.
The first service of this project is the ‘economical shopping map’ application.
Using the app, you can check the prices of fresh agricultural and livestock products and discount event information at stores within a 2-3 km radius based on the user's location.
If you search for a specific dish, such as kimchi stew, you will receive information that will help you efficiently and reasonably purchase the agricultural and livestock products needed for cooking.
This service will be operated on a pilot basis in five locations: Suwon, Gyeonggi Province, Wonju, Gangwon Province, Cheongju, North Chungcheong Province, Jeonju, North Jeolla Province, and Changwon, South Gyeongsang Province.
The government plans to introduce ‘AI national tax information counselor’ and national heritage AI interpretation service within the year. Next year, results on small business start-up and management consulting, provision of food and drug safety information, and voice phishing prevention will be revealed.
AI outlook — possibilities, not facts
A security-specific AI model will be completed within 10 months and distributed to governments and companies.
Likely · Within months
The ‘economical shopping guidance’ service will be expanded nationwide after pilot operation.
Possible · Within months

KT announced that its self-developed AI model routing technology 'Auto Model Router' ranked second in the comprehensive evaluation of accuracy and cost in Rice University's 'Router Arena' public leaderboard. This technology automatically selects the optimal AI model by analyzing the task type, difficulty, and knowledge field of the request, and is planned to be applied to the token factory to optimize service quality and cost.

North Chungcheong Province announced that it has established and started operating a 'Natural Disaster Situation Management System' that manages disaster situation information in one place. Compared to the existing system, report dissemination and follow-up actions were systemized and efficiency was improved by using AI to automatically input reports.

Samsung Electronics participated in the 2026 Mexico K-Expo held in Mexico City and connected products such as smartphones, TVs, and monitors based on Galaxy AI, Vision AI Companion, and SmartThings with AI technology to help visitors enjoy a wealth of Korean Wave content.

Microsoft co-founder Bill Gates said in an NBC interview that if AI falls into malicious hands, it could lead to the deaths of a billion people, and urged the U.S. government to pass legislation to establish safety measures and surveillance systems. He emphasized that self-regulation alone is not enough, and raised the need for international regulation by citing the recent OpenAI hacking incident into the Australian public health service and the Yemeni Houthi rebels' attempt to misuse the AI model Claude, but the United States dismissed this at the UN Security Council.

KT unveiled 'Agentic On', an agentic AI platform that connects AI agents and supports task automation using the company's existing systems and data. Multi-step tasks such as document review, approval requests, and system input are handled by multiple agents, and security and authority management functions are also included.

Robert O'Callaghan, a researcher at Google's DeepMind, announced his resignation, saying that AI is developing too quickly, and that religious beliefs and ethical concerns about technology were the reasons for his resignation. He explained that his work involved improving tools to accelerate AI chip design, and that the AI speed control debate was growing within the industry.