AI Export Controls Expand to Include AI Models, Raising Concerns for South Korea

The U.S. government's export controls on artificial intelligence (AI) are expanding beyond advanced semiconductors and graphics processing units (GPUs) to include AI models themselves. Citing national security concerns, the U.S. has directed foreign nationals to refrain from using Anthropic's latest AI models, 'Preamble 5' and 'Mythos 5.' In response, Anthropic has temporarily suspended access to these models. The key takeaway for South Korea is not merely the suspension of a single company's services, but the fact that the usage rights for completed AI models have now become a matter of national security control.

The essence of this situation is not a debate over technical flaws in specific models. The U.S. government has raised concerns about potential circumvention of safety measures and cyber misuse, while Anthropic has countered that these actions are excessive. However, a more significant shift is that the government is now directly determining 'who can use cutting-edge AI.' The criteria for accessing AI have shifted from chips and equipment to model usage rights, cloud utilization qualifications, and user assessments.

South Korea is not immune to this trend. Domestically, SK Telecom has joined Anthropic's Project Glasswing, securing early access to the AI model 'Claude Mythos' for detecting security vulnerabilities. This indicates that advanced U.S. AI is already linked to domestic security operations. This incident illustrates that such access rights could also be restricted under U.S. security policies. The challenge lies not just in adopting AI but in ensuring its stable and continuous use.

Of course, the conclusion cannot be to avoid using U.S. AI altogether. The highest-level AI models, cloud infrastructure, and security research ecosystems are predominantly driven by American companies. The issue is not collaboration but rather unguarded dependence. Relying on closed AI systems from foreign companies for core operations without alternative pathways could lead to new vulnerabilities in the supply chain rather than innovation.

To address this, sectors such as public services, finance, telecommunications, manufacturing, and security must assess potential disruption scenarios. If the interruption of specific AI services affects customer service, security detection, financial assessments, or manufacturing process management, it poses a national and industrial risk. The criteria for adopting AI should evolve beyond performance and cost to include 'can operations continue if services are interrupted?'

The government should also institutionalize the U.S.-South Korea AI trust partnership channel. Businesses and the public sector must revise procurement contracts to include provisions for notification of access interruptions, transition to alternative models, data transfer rights, and conditions for utilizing domestic backup systems. This will help ensure that core operations can continue despite external variables.

The Anthropic incident may end up being a temporary occurrence. However, the fact that the U.S. now views AI model usage rights as a national security issue remains significant. South Korea must deepen its collaboration with advanced U.S. AI while designing systems that can operate even when access is blocked. We have entered an era of export controls on AI, and South Korea's strategy must shift from 'adoption' to 'resilience.'

* This article has been translated by AI.