Who decides when AI is too dangerous?
TL;DR
The US government imposed export controls on Anthropic Fable 5 and the underlying Mythos 5 less than a week after Fable was released publicly. The order would have blocked foreign nationals, including Anthropic employees in the US, from accessing the models. Anthropic took both Fable and Mythos offline for everyone. The Verge reports that the move followed an alleged jailbreak found by Amazon researchers and escalated into a 90-minute ultimatum from the Trump administration.
Nauti's Take
Anthropic is getting hit by the downside of its own safety messaging. If a company spends years saying its models could become dangerous enough to require serious regulation, it should not be shocked when a government eventually reaches for the panic button.
The issue is not regulation itself. The issue is regulation by weekend phone call, which creates political risk instead of a credible safety framework.
Briefingshow
This is not just about one Anthropic model. If frontier models can be frozen through sudden export controls, political predictability becomes part of the infrastructure stack. For companies, AI risk is no longer just technical safety, compliance or cybersecurity.
It now includes geopolitics and access to power in Washington.
Sources