Who decides when AI is too dangerous?
TL;DR
The Verge reports that the US government placed export controls on Fable 5 and the underlying Mythos model shortly after public release. The rules would also block foreign Anthropic employees in the US from access. For Anthropic, that was hard to enforce cleanly in practice. Anthropic took both Fable and Mythos offline for everyone rather than rush a fragile access-control setup.
Nauti's Take
The most interesting question is not whether Fable 5 was truly dangerous. It is that one government move can apparently pull a frontier model offline worldwide.
That previews an AI world where release decisions no longer sit only with CEOs. PR language about responsible AI does not help much when nobody clearly explains which concrete threshold was crossed.
Briefingshow
AI safety looks less like an abstract ethics debate here and more like access control: who may see, test, use, or improve a model? When a government can restrict model access this quickly, control over AI shifts from product teams toward security and foreign policy.
Sources