The Ministry of Science and ICT will join OpenAI's trust-based cyber defense access program for governments and institutions, "GTAC." Korea, along with Japan, will be the first in Asia to join the program. This will allow the government to use high-performance AI models that OpenAI provides on a limited basis to study how AI models could be misused for cyberattacks and how to apply defensive technologies.
The Ministry of Science and ICT said on the 27th that Vice Minister Ryu Je-myeong met on the 26th with Jason Kwon, OpenAI chief strategy officer (CSO), and others to discuss responses to AI security threats and ways to ensure AI safety and trust. At the meeting, both sides agreed on the need for cooperation so the government and public institutional sector can use high-performance AI models safely, and finalized Korea's participation in GTAC.
The Korea Internet & Security Agency (KISA) will handle practical implementation. KISA plans to use OpenAI models for cyber defense research to assess defensive uses such as vulnerability detection, malicious activity analysis, and upgrading threat response systems. The Ministry of Science and ICT and OpenAI will continue working-level talks and will flesh out AI utilization plans that can be applied to domestic security operations.
AI safety evaluation was also a key agenda item. To reduce risks stemming from improving AI performance and wider use, the Ministry of Science and ICT proposed joint research between the AI Safety Research Institute and OpenAI and the establishment of a cooperation framework for safety evaluation. OpenAI was said to have noted it would review the Korean side's proposal. Vice Minister Ryu said the collaboration will help identify AI-based security threats in advance and raise Korea's response capabilities.
In the global AI industry, the trend of providing high-performance models for security defense on a limited basis is accelerating. OpenAI proposed a model of granting tiered access to GPT-5.5 and GPT-5.5-Cyber to vetted security personnel and institutions to support lawful defensive work such as vulnerability analysis, patch verification, and malware analysis.
Anthropic is also pursuing cooperation on checking core software vulnerabilities through "Project Glasswing" with AWS, Apple, Google, and Microsoft (MS). The government is also reviewing the possibility of joining this consortium, drawing attention to whether Korea's AI security cooperation will expand beyond OpenAI to big tech globally.