The U.S. government has begun treating artificial intelligence (AI) models similarly to advanced semiconductors and military technology. On June 12, Anthropic announced that the U.S. government directed it to halt foreign access to its latest AI models, 'Claude 5' and 'Mitos 5,' citing national security concerns. This order applies to all foreigners, including foreign nationals employed by Anthropic.
While this action may be a temporary response to safety concerns surrounding specific models, it signifies a more critical shift: the U.S. government now views AI models as strategic assets directly tied to national security.
This is not merely a case of corporate regulation; it signals that the AI competition has entered a new phase. Previously, the AI race was centered around semiconductor capabilities—who could secure more GPUs and build larger data centers. Now, the AI models themselves, which operate on these semiconductors, have emerged as core assets of national competitiveness. If strategic assets during the Cold War were nuclear weapons, the information age was defined by semiconductors, and the AI era is characterized by AI models, data, and AI agents as new strategic assets.
However, the decision has sparked considerable controversy. Anthropic argues that the jailbreak techniques cited by the government represent a limited vulnerability that can occur only under specific circumstances, and similar vulnerabilities can be found in other leading AI models. Critics within the industry contend that targeting specific companies and models raises fairness concerns.
The action reveals several contradictions. While the U.S. government has initiated regulations citing AI safety, it remains unclear whether the same standards will apply to other AI companies. Critics argue that blocking an entire commercial model based on a single vulnerability is excessive. If such precedents are established, AI companies may become hesitant to launch new models.
Another point of contention involves allied nations. While the U.S. emphasizes AI alliances, this measure effectively includes companies and research institutions from allied countries like the U.K. and South Korea in the access restrictions. Some experts criticize this approach, suggesting that it appears to prioritize controlling allies over countering China.
Yet, beyond these controversies lies a more significant issue: the question is not whether the U.S. government or Anthropic is correct. The essential reality is that AI has entered the realm of national security.
For South Korea, the situation is even more complex. Samsung Electronics and SK Hynix are leading the market for high-bandwidth memory (HBM), a crucial component in the AI era. The role of South Korean companies is vital for the operation of U.S. AI data centers and super-scale AI infrastructure. However, the most advanced AI models operating on top of this infrastructure are predominantly held by U.S. firms.
This incident starkly illustrates South Korea's reality. While we are a key supplier of essential components in the AI era, we rely on foreign AI models. If a specific country restricts access to AI models for political or security reasons, what choices do we have? This incident has brought that question to the forefront.
However, it would be premature to conclude that South Korea must develop its own GPT. There are differing perspectives within the AI industry. Companies like OpenAI, Google, Anthropic, and xAI invest tens of trillions of won annually. Competing on the same level to create world-class general-purpose AI is not realistically feasible for South Korea.
South Korea does not necessarily need to compete with the U.S. to create the world's top AI model. Not having developed a smartphone operating system does not mean South Korea is not an IT powerhouse. It is not essential to independently develop every core technology to maintain competitiveness.
However, creating the world's best AI and securing AI sovereignty are distinct issues. AI sovereignty does not stem from surpassing GPT; it arises from possessing the minimum AI capabilities necessary for a nation to choose and control its own technology when needed. Relying on foreign AI across critical sectors such as defense, finance, manufacturing, public administration, and healthcare could pose significant risks.
Therefore, South Korea should focus not on general-purpose AI competition but on industrial AI. Areas such as manufacturing AI, semiconductor AI, financial AI, defense AI, healthcare AI, and public AI present substantial opportunities. We possess world-class manufacturing competitiveness and industrial data, allowing us to achieve competitiveness comparable to the U.S. in specific sectors.
More importantly, the future competition will hinge on who can secure better data and build more powerful AI agents rather than merely who possesses larger LLMs. While LLMs may gradually become universal infrastructure like electricity or the internet, the creation of services and innovation in industries built on top of that infrastructure will be the responsibility of each nation.
The government must also shift its perspective on AI policy. If AI has been viewed as part of digital industry promotion, it should now be approached as a national strategic industry. Just as we nurture semiconductors, defense, and space industries, AI should be pursued as a long-term national project. Developing AI talent, securing data, building computing infrastructure, and creating industry-specific models should be approached from the perspective of national competitiveness.
Businesses should not perceive AI merely as a tool for improving productivity. AI is a technology that will transform the operating systems of all industries. The gap between companies that utilize AI and those that design it will only widen over time.
The fundamentals lie in technology. The principle is self-reliance. The common sense is that a nation cannot entirely entrust its future to foreign technology.
The Anthropic incident may end as a temporary event. However, the lesson for South Korea is clear: the AI supremacy competition has already begun. Semiconductors alone will not suffice. We must also be wary of illusions surrounding the world's top AI. What is needed now is the AI capability for a nation to choose and control its technology when necessary—essentially, AI sovereignty.
Whether South Korea remains a mere supplier of components in the AI era or becomes a technological powerhouse capable of designing its own future depends on the choices made today.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.