US reauthorises limited access to Anthropic's Mythos AI model for American partners after two-week national security block
Two weeks after forcing Anthropic to cut access to its most powerful AI model, the US government has restored it for a narrow set of American cyber-defence and infrastructure operators, while leaving foreign agencies locked out.
After an unprecedented two-week blockade, Washington has reauthorised a handful of US-based partners to use Mythos 5, the flagship model from Anthropic, citing national security conditions.
A swift clampdown
On June 12, Commerce Secretary Howard Lutnick forced Anthropic to immediately cut access to two cutting-edge models (Mythos 5 and a consumer-grade version called Fable 5), after the detection of undisclosed vulnerabilities. The move was framed as a national-security imperative. Anthropic, a leading AI company, complied and shut off all external access overnight.
- Commerce Secretary Howard Lutnick orders Anthropic to cut access to Mythos 5 and Fable 5
- Limited reauthorisation of Mythos 5 for US cyber-defence and infrastructure partners
Limited reauthorisation on June 26
On Friday 26 June, the Commerce Department reversed course partially, allowing a small circle of US-only partners to regain access. Anthropic said the approved users are "cyber-defenders and infrastructure operators" and that it is working to restore their access to Mythos 5 "as quickly as possible." The company did not specify the exact number of organisations involved.
Foreign partners remain cut off
The partial lifting does not apply to state cybersecurity agencies in Europe and Asia, who were also using Mythos under earlier agreements. They remain locked out, and the administration has given no timeline for a broader reopening.
Fable 5 still suspended
Fable 5, a consumer-oriented model derived from Mythos but with guardrails restricting capabilities in cybersecurity and biological or chemical risk, also had its access revoked on June 12. Its fate is still undecided, according to Anthropic.
Wider pressure on frontier models
Simultaneously, OpenAI indicated it has received strict instructions from Washington regarding access to its latest models, suggesting the tighter oversight extends beyond Anthropic. The moves confirm a more interventionist US posture on frontier AI, with national security screening now a condition for deployment.
