Anthropic launched its most ambitious artificial intelligence models on a Monday. By Thursday, the Commerce Department had severed global access.
On June 12, 2026, Secretary Howard Lutnick signed a directive ordering an immediate suspension of Claude Fable 5 and Mythos 5 for all foreign nationals, including the firm’s own non-US employees. Anthropic complied the same evening, disabling the models worldwide. The company described the incident as a “misunderstanding.”
The dispute focuses on a jailbreak that bypassed Fable 5’s safety guardrails. US officials assessed the breach could weaponize restricted cybersecurity features in the Mythos model. Amazon, the company’s largest investor and cloud partner, reportedly detected the vulnerability and notified regulators.
The order is narrow, targeting only the two newest releases. All other Anthropic models remain accessible. The firm intends to restore service once it addresses the government’s concerns. Still, applying software export controls within 72 hours of launch marks a sharp expansion of regulatory oversight.
Anthropic has long positioned itself as the industry’s safety leader. The incident now tests that reputation against federal enforcement, creating new exposure for major partners like Amazon as frontier AI deployment collides with national security policy.