Overview
- The U.S. Commerce Department issued a national‑security export directive that on June 12 ordered Anthropic to stop foreign‑national access to its newest models, Fable 5 and Mythos 5, and Anthropic disabled both models worldwide the same day.
- Government officials have pointed to a prompt‑based 'jailbreak' vulnerability as the technical reason for the order, but the actual directive has not been made public and agencies have given limited technical detail.
- Senator Mark Warner relayed that General Joshua Rudd told him Mythos breached nearly all classified systems within hours during testing, a reported claim that has intensified the security rationale but is not publicly corroborated with technical evidence.
- The global suspension cut off allied governments, research institutions and corporate users and prompted strong industry pushback; Anthropic publicly criticized the action and urged AI firms to consider relocating operations outside the United States.
- Policy experts say this episode exposes limits of applying traditional export controls to cloud AI, highlights enforcement problems for nationality‑based blocks, and is driving calls for EU and allied AI 'sovereignty' and new oversight frameworks.