He called for the US to adopt policy interventions to slow China’s AI development and said that all models — both open and closed — should go through mandatory safety testing before release