China may have accessed Mythos
TL;DR
The Verge, citing Semafor, says the White House export restrictions on Anthropic's Mythos were partly driven by fears that a China-linked group had accessed the model. If Mythos 5 or Fable 5 reached the Chinese government, the risk would include model theft by distillation, where a smaller 'student' model learns to mimic outputs from a stronger system.
Nauti's Take
This looks like a real security issue wrapped in a political power play. Until the White House and Anthropic show technical detail, both sides are asking the public to trust their framing: Anthropic minimizes the jailbreak angle, officials invoke national security, and China becomes the pressure amplifier.
The practical lesson is blunt: frontier AI launches now combine product rollout, export control, incident response, and geopolitical risk in the same release checklist.
Briefingshow
Frontier models are being treated like strategic infrastructure: access control, export law, employee eligibility, and national security now collide inside one product decision. For companies, user location, staff nationality, and model-access logs can become operational blockers overnight. The evidence is still thin, but the policy reflex is already visible.
Sources