“Uses high-stakes language like 'alarm' and 'existential dangers' to frame AI as a severe threat.”