en.Wedoany.com Reported - The Korea Institute for National Security Strategy (INSS) released a report stating that as the United States has placed frontier artificial intelligence (AI) under national security control, South Korea also needs to establish a national-level frontier AI risk assessment system and redesign its sovereign AI strategy.
In an issue brief published in early July, the institute analyzed that the "Mitose incident," where U.S. government export control measures led to service disruptions of Anthropic's models, has made overseas frontier AI access a core variable in national security. U.S. President Donald Trump signed an executive order on "Promoting Advanced AI Innovation and Security" on the 2nd of last month, targeting frontier AI models, including measures such as mandatory self-review within 30 days, classified benchmark testing led by the National Security Agency, and the establishment of an AI cybersecurity information exchange. The report noted that the Trump administration, which had maintained a stance of technological deregulation, has explicitly designated frontier AI as a national security control target, marking a shift in its AI management system from industrial policy to national security governance.
The background of the U.S. government's policy shift is the Mitose incident. In April this year, Anthropic stated that its "Claude Mitose Preview" model had acquired the cyber capability to autonomously detect zero-day vulnerabilities and generate attack code. Consequently, on the 12th of last month, the U.S. Department of Commerce, citing national security reasons, requested restrictions on access to "Mitose5" and "Fable5," which is intended for general users. Anthropic completely suspended services, citing the difficulty of distinguishing between overseas users and foreign users within the United States. After 18 days, the controls were lifted, and Fable5 officially resumed service on the 1st of this month.
The report cited Anthropic's decision to implement a broader service suspension than the government required as an example, pointing out that actual regulatory effects may vary depending on corporate discretion. It also analyzed that shortly after the service suspension, China's Zhipu AI launched "GLM-5.2," suggesting that gaps in specific models could create opportunities for competitor nations.
The report proposed South Korea's response measures, including establishing a national risk assessment system centered on cyber capabilities, rather than the computing power-based "AI Basic Act"; establishing AI security governance with the National Intelligence Service at its core; and redesigning a sovereign AI strategy that defines AI sovereignty as including access continuity. INSS Research Fellow Yang Ji-soo stated that future AI policies must consider not only what to develop but also how to ensure the continuity of access to frontier AI.
