US export controls on Mythos put South Korea's Project Glasswing membership in jeopardy

The US administration has imposed export controls on Mythos, the top-tier AI model developed by Anthropic, dealing a blow to South Korean government agencies and companies that had only just joined the global AI security coalition known as Project Glasswing.

Anthropic announced Thursday that the US government, acting on guidance from national security authorities, had issued export control directives cutting off foreign nationals from accessing Mythos 5 and Fable 5.

The restrictions apply broadly — covering users connecting from abroad, foreign nationals residing in the United States, and even Anthropic's own foreign employees.

The move directly affects South Korean government bodies and companies that had recently begun participating in Project Glasswing.

Project Glasswing is a global cybersecurity cooperation initiative that Anthropic launched alongside its announcement of Mythos. Concerned that the model could be exploited by malicious hackers, Anthropic designed the program to give vetted companies and institutions early access to Mythos so they can identify and defend against software vulnerabilities.

Anthropic had significantly expanded Glasswing membership on June 2, adding roughly 150 new institutions across 15 countries. Among the South Korean organizations that joined were Samsung Electronics, SK Hynix, SK Telecom and the Korea Internet & Security Agency (KISA).

The Korean participants had gained access to Mythos through their Glasswing membership, marking their first step into global AI security cooperation. But barely 10 days after joining, they were blindsided by the US administration's export control measures. While access credentials had been granted, the organizations had not yet reached the stage of actively using the model.

"We have been in ongoing communication with Anthropic, so we are calmly working to clarify the facts," a Ministry of Science and ICT official said, adding that the government is discussing a response through a consultative body led by the National Security Office.

South Korean companies are also in contact with Anthropic as they work to assess the situation.

The episode has renewed attention in South Korea on the importance of building homegrown AI capabilities — what is often called "sovereign AI."

Ha Jung-woo, former presidential AI planning secretary at Cheong Wa Dae, weighed in on his personal social media account, calling the situation "what happens when a country becomes technologically dependent on AI in an era where AI capabilities determine national competitiveness." He added that "this kind of thing can keep happening at any time," and stressed that "while cooperating globally, countries must also have their own capabilities for when things go wrong."

An Anthropic official said the company believes the issue stems from a misunderstanding and pledged to restore service "as quickly as possible."