China may have accessed Mythos
TL;DR
Semafor reports that US export limits on Anthropic were partly driven by fears that a China-linked group had accessed Mythos, one of Anthropic’s most powerful models. The order covered Mythos and its consumer version Fable 5, limiting access to US citizens. Anthropic instead pulled the models from the market. The China angle is not confirmed. David Sacks publicly focused on a jailbreak warning, while Anthropic says China was not raised in the export-control talks.
Nauti's Take
This shows how fast frontier models can shift from product issue to state-security problem. With cybersecurity models, even alleged access is enough to scramble export logic, market availability, and geopolitical framing.
The messaging is PR-heavy on all sides: the government, advisers, and the company are each pushing the version that creates the least damage for them.
Briefingshow
This turns frontier AI from a product question into a national security question. The issue is not only who can use a model, but whether even temporary access can transfer capabilities into other systems. That is why AI export controls are starting to look less like software policy and more like chip and security policy.
Sources