[Editorial] Urgent Need to Secure Self-Reliant Capabilities in the Face of U.S. Advanced AI Export Controls

The U.S. government has completely barred foreign access to Anthropic's latest AI models, Claude Fable 5 and Claude Mythos 5, which have drawn attention for their cybersecurity capabilities. The ban applies not only to users outside the United States, but also to foreign nationals inside the country and even the company's own foreign employees. The administration cited national security. In effect, it is a declaration that advanced AI will be managed like a strategic weapon.
The United States Department of Commerce formalized the measure on the 12th local time through guidance issued in the name of Secretary Howard Lutnick. The United States has previously restricted exports of semiconductors and graphics processing units, but controlling AI models themselves is unusual. It shows that AI competition is expanding beyond chips to models, algorithms, and ultimately the realm of national security.
The move is also expected to disrupt plans by Korean institutions participating in Project Glasswing, a global AI security consortium led by Anthropic. Samsung Electronics, SK Telecom, and Korea Internet & Security Agency (KISA) are among the domestic companies and institutions known to be involved. The episode shows how a plan to use U.S. technology to counter AI-based cyber threats could be shaken overnight.
This development raises once again the question of why Sovereign AI is necessary. AI is no longer just a service that makes daily life more convenient. It is a core infrastructure that shapes the economy, industry, defense, and security. As dependence on foreign technology grows in a world where a handful of countries and companies dominate cutting-edge AI, vulnerability to supply disruptions and regulatory changes becomes inevitable.
The world is already entering an 'AI Cold War.' The United States has begun treating cutting-edge AI as a strategic asset, while China is responding with a full-scale national effort. Controls over AI models, data, and computing infrastructure are likely to tighten further. If South Korea hesitates now on investment in technology, talent, and infrastructure, it could end up not as an AI powerhouse, but as a mere consumer of technology.
The government has also launched development of its own foundation model, aiming for self-reliance by 2027. But the pace must accelerate. Model development alone is not enough. The country must secure computing infrastructure such as data centers and build an ecosystem where domestic AI can be used in public, defense, and industrial settings. It also needs a multi-supply-chain system that moves away from relying on a single AI for cybersecurity and instead uses multiple overseas and open-source models.
Global AI competition is increasingly solidifying into a U.S.-China duopoly, with American companies maintaining the highest performance and Chinese firms close behind. South Korea is often counted among the top three countries with its own AI model, but the gap remains wide compared with those two powers.
Sovereign AI is not a slogan, but a survival strategy. This U.S. move is a reminder of the risks faced by countries that rely solely on others' technology in an era of technological hegemony. Rather than speaking of crisis too late, South Korea must mobilize national capabilities now to build the foundation for AI self-reliance. There is not much time left.