China may have accessed Mythos
TL;DR
Semafor reports that White House export restrictions on Anthropic Mythos were partly driven by fears that a China-linked group had accessed the model. That China angle is unconfirmed: The White House has not verified it, and David Sacks publicly focused on alleged jailbreak risks instead. If China actually obtained access to Mythos 5 or Fable 5, the national security concern would be real, not cosmetic.
Nauti's Take
The story is explosive, but still lightly sourced. It should not be framed as a confirmed China breach.
The bigger signal matters more: Anthropic describes these models as too powerful for broad release, while access failures and political interventions keep surfacing. Frontier AI companies cannot lead with safety language alone; access control has to hold up in practice.
Briefingshow
This shows how quickly frontier models shift from product launches to strategic infrastructure. When access, jailbreaks, and export controls collide, the issue is no longer just safety messaging; it is control over capabilities that can be copied, probed, and turned into geopolitical leverage.
Sources