Who decides when AI is too dangerous?
TL;DR
The Trump administration placed export controls on Anthropic’s new Fable 5 model and the underlying Mythos 5 shortly after launch. The order also covered foreign nationals in the US, including Anthropic employees, so Anthropic pulled both Fable and Mythos offline for everyone. According to The Verge, the trigger was a possible jailbreak surfaced by Amazon red-team research and a reported 90-minute government ultimatum.
Nauti's Take
Anthropic spent years arguing that frontier AI could become dangerous. Now it is seeing how ugly regulation gets when it is not clear, transparent and technically grounded.
That does not absolve Anthropic: if you describe models as potential cyber weapons, governments will get jumpy. The problem is the operating mode: 90-minute pressure and vague access rules are not a safety framework.
Briefingshow
The issue is not only whether Fable 5 was dangerous. The bigger question is who gets to draw the line: safety researchers, AI companies, regulators or political relationships. If export controls become an emergency lever, companies will have to model political risk alongside model risk.
Sources