Key Takeaways

  • The Trump administration ordered Anthropic to restrict Claude Mythos and Fable 5 access to US nationals, citing national security concerns.
  • SK Telecom’s participation in Project Glasswing and Amazon’s vulnerability findings intensified government scrutiny.
  • The move highlights growing tensions around safeguarding advanced AI and raises questions about how global collaboration can continue under tightening controls.

US officials learned that Anthropic had granted SK Telecom access to Claude Mythos within Project Glasswing, a program designed to provide limited access to an AI model highly skilled at identifying software vulnerabilities. The Trump administration’s move to impose export controls on Anthropic’s most powerful AI technology followed this dispute, driven by concerns over SK Telecom’s alleged ties to China.

SK Telecom has emphasized that claims about ties to China lack verified facts, and the US government letter demanding restricted access to Claude Mythos did not explicitly reference the South Korean company. However, for policymakers sensitive to supply chain exposure and data flows, perceived indirect ties prompted strict regulatory action.

Amazon researchers also flagged vulnerabilities in Fable 5, a highly safeguarded version of Claude Mythos that Anthropic released to the public on June 9. The Amazon report to the White House suggested it was possible to circumvent some of Fable 5’s guardrails and access advanced cyber capabilities. Anthropic and outside cybersecurity experts countered that these risks are not unique to Claude. Ultimately, the confluence of these events led the White House to determine it could not trust Anthropic to safeguard its most advanced AI technology.

This pattern of regulatory intervention aligns with challenges analysts at the GAO highlight regarding federal oversight of complex digital systems. Agencies face a recurring dilemma of balancing defensive governance structures with the rapid innovation and substantial economic value presented by advanced AI systems.

On Friday, the Trump administration ordered Anthropic to revoke access to Mythos and Fable 5 for all foreign nationals, including immigrants living inside the United States. Rather than attempt to authenticate users’ nationality across thousands of customers while upholding privacy requirements, Anthropic disabled access to the models entirely. Anthropic and the White House remain in negotiations to bring Claude Mythos and Fable 5 back online.

Export control frameworks were originally built for hardware, but regulators are stretching these policies to cover AI. Researchers at IEEE note this trend heavily impacts the intersection of networking standards and digital sovereignty, raising questions about how compliance can function if AI models continue to evolve at their current pace.

Earlier this month, SK Telecom, South Korea’s largest wireless carrier, became one of roughly 150 companies granted access to Mythos. Anthropic expanded the Project Glasswing program following close collaboration with outside experts. While Anthropic maintained that SK Telecom’s access and the vulnerabilities identified by Amazon were separate issues, the simultaneous scrutiny amplified government intervention.

Historical interactions between SK Telecom and entities like China Unicom, including past joint ventures and investments in UNISK, often resurface in policy discussions during moments of geopolitical tension. Regulators interpret supply chain risks aggressively, and past international telecom investments undergo rigorous scrutiny even when formal strategic relationships have concluded.

Broader proposals to bar US telecom carriers from interconnecting with China Unicom illustrate a tightening environment for cross-border communications infrastructure. China Unicom warned that such measures could disrupt global networks, adding another layer of uncertainty to international data flows.

Industry observers studying geopolitical technology shifts, including analysts at Brookings, note that AI has entangled with national security policymaking rapidly. The Anthropic restrictions demonstrate how global collaboration becomes difficult when models, datasets, and supply chains are treated as sensitive assets.

The regulatory action reflects the strain between global AI development and national security controls. Organizations must navigate governance hurdles that dictate where they invest, which partnerships they pursue, and how they manage international data pathways under modern export frameworks.