On June 12, Commerce Secretary Howard Lutnick signed an order requiring Anthropic to obtain US government approval before any foreign national could use Fable 5 or Mythos 5. Bloomberg, reporting it out on June 19, called it an unprecedented use of export control law, extending statutes built for transfers of sensitive technology to cover the mere act of querying a commercial AI model.
Two events, roughly a week apart, set it off.
The first was Project Glasswing, Anthropic’s vulnerability-detection initiative, which on June 2 expanded to roughly 150 organizations across more than 15 countries. Among the South Korean participants were SK Hynix, Samsung, and SK Telecom, the country’s largest wireless carrier. In early June, per Wired’s June 17 reporting, SK Telecom was added to the program and then quietly removed at the White House’s request over alleged ties to China. SK Telecom’s own annual report shows a China footprint of roughly $1.9 million in 2024 revenue and seven employees. The company told a Korean newspaper it had “no ties to China” and that the “anonymous insider’s remarks in foreign media lack verified facts.”
The second event, per Tom’s Hardware, came three days after Fable 5’s June 9 public release. Researchers at Amazon, sitting on a roughly $13 billion cumulative stake in Anthropic, prompted Fable 5 to read a codebase and fix its flaws, a clean guardrail bypass. Amazon CEO Andy Jassy reportedly raised the finding with administration officials directly. Three days later the Commerce order landed.
Anthropic, in a company statement, described the underlying problem as “a limited number of isolated circumvention cases.” Roughly 100 cybersecurity professionals, including former Facebook security chief Alex Stamos and Luta Security’s Katie Moussouris, signed an open letter arguing that Mythos and Fable “are not uniquely good” at weaponizing software flaws, the technical premise on which the export logic rests.
The diplomatic recovery began almost immediately. On June 17, the same day Wired published, Anthropic opened a Seoul office and signed a memorandum of understanding with South Korea’s Ministry of Science and ICT. The lawyers and developers Bloomberg spoke with raised the open question now circulating in Washington and Seoul: whether Commerce can dictate who gets to query a commercial AI system on national-security grounds, and what precedent answers yes.
The structural read is harder than the news. Export control statutes were written for objects that cross borders. The Lutnick order treats inference itself as the border, which makes every API key a potential license application and every foreign customer a compliance question routed through Washington. The Seoul MoU is what that world looks like in practice.
Sources
- https://www.bloomberg.com/news/articles/2026-06-19/lutnick-s-anthropic-crackdown-claims-new-power-over-ai-models
- https://www.wired.com/story/sk-telecom-anthropic-mythos-export-controls/
- https://www.anthropic.com/news/fable-mythos-access
- https://techcrunch.com/2026/06/02/anthropic-scales-claude-mythos-to-critical-infrastructure-in-15-countries/
- https://www.tomshardware.com/tech-industry/artificial-intelligence/sk-telecom-named-as-the-korean-carrier-at-the-center-of-anthropics-mythos-export-controls