What happened
On the evening of June 30, 2026, Commerce Secretary Howard Lutnick notified Anthropic by letter that the Department of Commerce had lifted the emergency export controls imposed on June 12, 2026 on Claude Fable 5 and Mythos 5. The controls had barred all foreign nationals from accessing both models, citing national security risks from a reported jailbreak vulnerability. The June 26 partial lift (Mythos 5 to ~100 trusted partners) and the June 30 full lift (both models, globally) are a new development: they come with binding conditions Anthropic agreed to as the price of restoration — proactively detecting and addressing security risks, collaborating with the US government on protocols and standards for Mythos, Fable, and future models, and informing the government of any malicious activity. Mythos 5 access is also being expanded through the 'Glasswing' programme for vetted cybersecurity researchers.
Why it matters
This is the first confirmed instance of US emergency export controls being applied to — and then conditionally lifted from — a specific commercial frontier AI model. It establishes that advanced AI models and their APIs can be treated as controlled technology under national-security export-control authorities, sets a precedent for government-mandated pre-release security review of future frontier models, and imposes ongoing compliance obligations (monitoring, reporting, government coordination) on Anthropic as a condition of global distribution. The entire episode — June 12 ban through June 30 full lift — signals a new managed-access paradigm for the most capable US AI systems.
Action needed
AI developers and deployers relying on frontier-model APIs from US labs should review their contracts and supply-chain risk plans for government-imposed access suspensions. Anthropic customers outside the US should confirm restoration of access. Watch for the August 2026 executive-order deadline requiring Commerce/ONCD to formalise standards for vetting future frontier model releases.