The US government on June 12 banned foreign access to Mythos5 and Fable5, the cutting-edge AI models developed by Anthropic, blocking not only overseas users but also foreign nationals and foreign employees inside the United States. The sweeping restriction came just three days after Fable5's public release. As the first export control ever applied to an AI model, it signals that AI is now regarded — alongside semiconductors and military technology — as a domain directly tied to national security.
The trigger was the discovery of a potential "jailbreak" — a method to bypass Fable5's built-in safety filters. The model was designed to block instructions that could be exploited for cyberattacks or the development of biological and chemical weapons, but researchers reportedly found ways around those guardrails. Washington concluded that malicious actors could exploit the vulnerability to pose a serious threat to national security. Some foreign media outlets also cited concerns that entities linked to China may have accessed the model as part of the regulatory backdrop. When Anthropic declined a White House request to suspend the service, the administration escalated to export controls.
The severity of the response reflects the raw power of the Mythos and Fable models. Both are assessed to possess expert-level cybersecurity capabilities — able to identify software vulnerabilities, analyze security architectures and predict potential attack vectors. While that makes them valuable for auditing critical infrastructure and bolstering cyber defenses, the same capabilities could enable devastating attacks if misused. The era in which AI attacks and AI defends is becoming reality.
The concern for Korea is that its AI competitiveness lags well behind the front runners. According to AI performance evaluator Artificial Analysis, Fable5 ranks first in the world, followed by OpenAI and Google. Behind them, Chinese firms — Alibaba, DeepSeek, Moonshot AI and Xiaomi — have clustered in the middle tier. Stanford University's Institute for Human-Centered AI has ranked Korea third globally, behind the United States and China, but its actual competitive standing remains a considerable distance from the leaders. The fact that Korea can sustain independent model development is encouraging, but much ground remains to be covered.
Korea would do well to study how China has grown. Despite US semiconductor restrictions that have made it difficult to secure the most advanced chips, China has cultivated a thriving AI ecosystem through intense competition among numerous companies, low prices and open-source models — building competitiveness on its own terms. Korea must create an environment where more companies can compete, and expand its data and computing infrastructure. It should also move quickly to nurture an AI ecosystem that plays to its strengths in manufacturing, public services and the Korean language. AI technology now shapes both national competitiveness and national security.
