On the 12th (local time), after the U.S. government issued guidance restricting overseas access to Anthropic's latest artificial intelligence (AI) models "Mythos 5" and "Fable 5," the need for sovereign AI (AI capabilities that a country can develop, operate, and control on its own) is being highlighted again in Korea. That is because the prospect that cutting-edge AI models can be controlled as a national strategic asset has become a reality.
The move is said to stem from concerns that the models could be "jailbroken," a so-called bypass of prompt controls. Anthropic said the matter arose from a misunderstanding, but it first blocked all access to comply with the guidance.
Experts in the science and technology community see this situation as a direct warning for Korea as well. It showed that in a landscape with heavy reliance on overseas foundation models, AI access can be restricted by a particular country's policy decisions. Foundation models are large, general-purpose models that underpin a range of AI services such as chatbots, translation, search, coding, and medical and scientific analysis.
Choi Ki-young, former Minister of the Ministry of Science and ICT (professor, Department of Electrical and Computer Engineering, Seoul National University), said, "There have long been many concerns that AI could be treated as an important national strategic asset and that access by other countries could be controlled, and this incident appears likely to further heighten that vigilance."
He added, "High-performance AI models can be used for both cyberattacks and defense, and it cannot help but be a major threat to other countries if one nation monopolizes such models," and said, "This should be an opportunity for Korea to accelerate efforts to develop top-tier, homegrown foundation models."
However, developing an indigenous model requires large-scale compute resources, talent, data centers, and investment. The former Minister said, "For Korea, which lacks talent, graphics processing units (GPUs), data centers, and investment, it is not easy," adding, "Industry, academia, research, and government must cooperate more closely."
Kim Nam-kuk, professor of convergence medicine at Ulsan University College of Medicine and Asan Medical Center, predicted that this step will further spur the global competition to develop AI. Kim said, "Paradoxically, I expect it will become a boomerang that weakens the U.S.-centric global AI ecosystem in the long term," adding, "Beyond the bipolar system of the United States and China, it will be reorganized into a new equilibrium, such as a tripolar structure in which excluded countries band together."
He went on to say, "Right now, the domestic sovereign AI development strategy tends to insist too much on building fundamental technologies," and added, "For security, we should first develop high-performing AI from an application-technology perspective, and we need a multifaceted response through institutional support for big data discovery, data use and compensation, and AI infrastructure investment."
Lee Sang-yup, distinguished professor in the Department of Chemical and Biomolecular Engineering and vice president for research at KAIST, said it is more important for Korea to secure AI capabilities that work without overseas access in key fields, rather than trying to immediately catch up with U.S.-level, ultra-large models in every area.
Lee said, "Ultimately, it is good to develop sovereign AI by投入 massive resources, but resources are not always so abundant," adding, "Even if usage controls are imposed by AI-leading countries, we must have models in place that operate well for defense, health care, research, and administration."