Who decides when AI is too dangerous?
TL;DR
The Verge frames the Anthropic, Fable 5 and Mythos episode as the first major stress test for the Trump administration’s new AI regulation posture. After a reported jailbreak concern surfaced via Amazon, the US government imposed export controls that also blocked foreign Anthropic staff in the US from accessing the models. Anthropic took Fable and Mythos offline for everyone because it said it could not reliably separate access by nationality fast enough to comply.
Nauti's Take
Anthropic has spent years arguing that powerful AI systems may become dangerous. Now it is seeing how messy regulation gets when it arrives as a political panic button instead of a serious framework.
The problem is not government intervention itself. The problem is a process where nobody can clearly say who decides, what evidence counts and what legal standard applies.
Briefingshow
The case shows how quickly AI safety can shift from technical risk assessment to geopolitical control. If governments can force model shutdowns on short notice, they need clear criteria, competent oversight and reviewable procedures. Otherwise the result is not safety, but uncertainty for researchers, companies and international teams.
Sources