“Uses urgent, security-threat language like 'rogue' and 'explosive' to frame AI safety as a systemic crisis.”