Korea Scrambles Over 'Mythos' Export Controls... Sovereign AI Gains Momentum

The Challenge of Narrowing the Gap in Capital-Intensive AI

Growing Need for Domestic AI in Defense, Security, and Public Sectors

The U.S. government has begun export controls on Anthropic's high-performance artificial intelligence (AI) model "Mythos," prompting the South Korean government to urgently prepare countermeasures. There is growing concern that if such government controls continue to expand, AI models could become strategic goods similar to semiconductors.

On the morning of the 15th, the Ministry of Science and ICT discussed the U.S. export controls on Mythos during its work meeting. Because the U.S. government has exercised direct control over a high-performance AI model, the South Korean government is now busy verifying the details and discussing countermeasures. A Ministry official stated, "We are currently confirming the facts regarding the U.S. government's export control on Mythos and are discussing specific response measures." An SK Telecom official also said, "We are monitoring the situation in communication with Anthropic following the U.S. government's actions."

On June 12 (local time), Anthropic announced that, in accordance with instructions from U.S. national security authorities, it would fully suspend foreign nationals’ access to "Mythos 5" and "Fable 5" via new export control guidelines. These measures apply not only to overseas users but also to foreign nationals residing in the U.S. and even foreign employees of Anthropic. Immediately following the issuance of these guidelines, Anthropic suspended the Fable 5 and Mythos 5 services to comply with the regulations.

Industry insiders are concerned about the impact of the Mythos export controls, given that AI models have now been elevated from tools of national competitiveness to strategic goods subject to state control, with significant implications for national security. If access controls are expanded to widely used models such as ChatGPT, Claude, or Gemini in the future, it could pose a major threat at both the corporate and national levels.

Kim Seungjoo, professor in the School of Cybersecurity at Korea University, said, "The U.S. government blocked the export of Mythos because it considered it critical to national security and made its decision from a strategic goods control perspective." He added, "In the field of national security, urgent research is needed to improve security using sovereign AI."

In particular, Mythos is not a general generative AI, but is known for its outstanding vulnerability detection and attack path analysis capabilities, making access acquisition a high priority for many countries. "Project Glasswing" is a security consortium established by Anthropic in April when it announced the development of its latest model, "Claude Mythos." The aim is to use Mythos to detect cyber vulnerabilities across various industries. Institutions and companies participating in the consortium are granted the right to use Mythos. In South Korea, Samsung Electronics, SK hynix, and SK Telecom have been included as participants in Project Glasswing. The Ministry of Science and ICT also secured access through the Korea Internet & Security Agency (KISA), but this export control measure has inevitably disrupted AI-based cyber threat responses.

There are growing calls for South Korea to enhance its own security capabilities and develop measures for vulnerability analysis using the latest AI models in order to respond to AI-driven cyber threats. With the recent emphasis on domestic AI model adoption in sensitive areas such as defense, the public sector, and national security, this could serve as an inflection point for sovereign AI. The government is also expected to accelerate its initiatives to localize the AI ecosystem, including large language models (LLMs) and AI semiconductors, centered around the "Independent AI Foundation Model" (DOKPAMO) project.

Lee Jinwon, Chief Technology Officer (CTO) of HyperAccel, stated, "In the case of some Chinese AI models, foreign use is already restricted, so the U.S. export controls were to some extent anticipated." He emphasized, "AI is a capital-intensive field that, unlike conventional software, requires significant infrastructure such as GPUs. To narrow the gap with the U.S. and China, the government must not only secure its own technologies through projects like DOKPAMO, but also proactively seek and invest in innovative solutions for establishing sovereign AI."

Choi Byungho, professor at the Human-Inspired AI Research Institute at Korea University, suggested, "Going forward, the U.S. government is highly likely to turn AI models into strategic assets and use them as bargaining chips in negotiations." He added, "While we should maximize the use of models from Anthropic, OpenAI, and Google for security, it is essential to advance our own AI models for cyber defense."

This content was produced with the assistance of AI translation services.