Who decides when AI is too dangerous?
TL;DR
Anthropic took Fable 5 offline soon after its public launch. The model is the guarded consumer version of Mythos 5, which Anthropic has framed as a much riskier frontier system. The US government imposed export controls barring foreign nationals from Fable and Mythos, even Anthropic employees in the US. Anthropic said it could not separate access cleanly fast enough.
Nauti's Take
This looks like a safety debate with a heavy layer of power politics. Anthropic spent years telling the world its strongest models could become dangerous; now a government is using that story as leverage.
A 90-minute ultimatum is still not serious AI oversight. If frontier models need controls, the process needs technical review, clear thresholds and deadlines that do not depend on who picked up the phone fastest.
Briefingshow
This is bigger than Anthropic's fight with Trump. When a model can disappear overnight through export controls, companies have to price political risk into AI contracts, staffing and security work. The gap is now visible: labs sell voluntary safety language, while governments still lack a predictable review process for frontier models.
Sources