“Uses high-stakes language like 'rogue' and 'exploding' to frame AI incidents as a rapidly escalating crisis.”