Who decides when AI is too dangerous?

An AI company released a new public model with strict safety guardrails, but the U.S. government imposed export controls on both that model and its underlying version, citing national security concerns. The company then took both models offline because it was worried it could not comply with restrictions that prevented even its own foreign national employees from accessing them. This situation raises questions about how the U.S. government will decide when AI is too dangerous and whether that regulatory approach will be based on genuine safety concerns or used as leverage against companies and people the administration opposes.