How Anthropic’s Seoul Expansion Clashed With a US Export Order in One Week

The AI Safety MOU

The centerpiece of the announcement was a Memorandum of Understanding (MOU) with South Korea’s Ministry of Science and ICT (MSIT) to advance AI safety and cybersecurity cooperation . Under the agreement, the Artificial Intelligence Safety Institute (AISI) under the Electronics and Telecommunications Research Institute (ETRI) and Anthropic agreed to cooperate on AI safety, including evaluating model safety in the Korean language . In cybersecurity, MSIT and Anthropic will jointly identify AI vulnerabilities and share expertise on cyber threats .

Crucially, Korean authorities had already secured access to Anthropic’s cybersecurity AI model Mythos on June 3, 2026, through participation in Project Glasswing alongside major South Korean companies . This gave South Korea access to frontier AI safety tools even before the public Seoul announcement .

The Export Control Directive: A Global Shutdown of Fable 5 and Mythos 5

Just five days before the Seoul office opening, a very different dynamic unfolded. On June 12, 2026, the U.S. Commerce Department issued an export control directive ordering Anthropic to immediately suspend access to its two most advanced models — Claude Fable 5 and Claude Mythos 5 — for any foreign national, whether inside or outside the United States, including Anthropic's own foreign-national employees . The directive was signed by U.S. Commerce Secretary Howard Lutnick .

Why the Models Were Disabled Globally

Because Anthropic could not reliably verify every user's nationality in real time at the system level, the only path to compliance was to disable both models for all customers worldwide . Anthropic released an official statement saying, “The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance” . The company described it as a “misunderstanding” and said it was working to restore access .

The Stated Trigger

The Commerce Department cited a claimed jailbreak with national security implications as the trigger for the directive . The Cloud Security Alliance described the episode as “the first time the U.S. government has directly compelled an AI company to revoke access to specific deployed model versions based on user nationality” . The models had only been available to the public for three days — Fable 5 was released on June 9 .

Anthropic’s Disputed Rationale

Anthropic reportedly disputed the government’s rationale, arguing that the cited demonstration was narrow rather than a universal compromise of the models . The company warned that applying such a standard broadly could severely constrain the deployment of frontier AI models .

How the Two Events Relate

The Seoul office opening and the export-control shutdown were not directly causally linked, but their timing created an undeniable tension.

• The Seoul expansion was a long-planned strategic push into Korea’s AI ecosystem, focused on safety partnerships and commercial growth .
• The export-control order was a sudden national-security action that forced Anthropic to restrict access to its most advanced capabilities — including to the very Korean partners it was simultaneously courting .

Notably, the Seoul announcement and the MSIT MOU focused on AI safety, cybersecurity cooperation, and Korean-language model evaluations — not specifically on access to Fable 5 or Mythos 5 . South Korea had already secured access to the Mythos cybersecurity model through Project Glasswing before the shutdown was reported .

Awkward Commercial Reality

Despite this, the export-control directive created an awkward commercial reality. Anthropic was expanding partnerships with Korean enterprises, researchers, and government institutions while also disabling its most advanced models globally because of restrictions on foreign-national access . The directive applied to all foreign nationals, including Anthropic's own non-citizen employees , making compliance especially challenging for a company with a global workforce.

The Broader Implications

This episode marks a pivotal moment for AI governance. The Seoul expansion and the export-control order together illustrate the central challenge facing frontier AI companies: balancing global commercial ambitions and safety partnerships with a rapidly evolving patchwork of national security controls.

South Korea's proactive approach — securing access to frontier safety tools through Project Glasswing while formalizing an MOU on AI safety — stands in contrast to the U.S. government's unilateral export-control action. The question now is whether other nations will follow South Korea's path of partnership or whether export restrictions will increasingly define the global AI landscape.