The most advanced artificial intelligence has become a geopolitical battleground. The case involving SK Telecom, South Korea's largest telecom operator, and Anthropic, the startup founded by former OpenAI members, has highlighted how national security is reshaping access to frontier AI models. The story revolves around Claude Mythos, Anthropic's most powerful model, and its guarded version Fable 5, released to the public on June 9.
According to reports, the White House ordered Anthropic to revoke access to Mythos and Fable 5 for all foreign nationals, including immigrants in the US, after the government expressed concerns over SK Telecom's alleged ties to China. SK Telecom, which invested USD 100 million in Anthropic in 2023 and is part of the SK Group conglomerate, denied any connection to Beijing. However, its stake in a joint venture with China Unicom and the group's interests in semiconductors and energy fueled suspicions.
Sponsored Protocol
The White House decision was also accelerated by the discovery of vulnerabilities in Fable 5 by Amazon researchers. Although Anthropic and external cybersecurity experts argued the risks were not unique to Claude, the government deemed that Anthropic could not adequately safeguard the technology. Faced with the difficulty of restricting access based on nationality without violating privacy, Anthropic chose to disable the models entirely, sparking a dispute with the Trump administration.
The Project Glasswing program had been launched to give early access to Mythos to about 150 trusted organizations, including Samsung Electronics and the Korea Internet and Security Agency. SK Telecom was included after weeks of collaboration with experts and the US government. However, the program expansion alarmed the White House, which called for revoking SK Telecom's access earlier in June. Anthropic complied immediately, but pressure mounted after Amazon's report.
Sponsored Protocol
This incident highlights growing tensions in the AI ecosystem. On one hand, the need for global collaboration to develop safe models; on the other, the fear that dual-use technologies could fall into the wrong hands. The SK Telecom case is a wake-up call for Europe, which must accelerate the definition of clear AI export rules. As discussed in AI Under US Control, the issue is not just technical but political. Meanwhile, companies like Pramaana Labs are working on formal verification methods to increase trust in AI systems.
For further context on the geopolitical landscape, consult the Wikipedia page on SK Telecom.
Source: https://www.wired.com/story/sk-telecom-anthropic-mythos-export-controls
