“Words like 'hacked' and 'growing unease' frame AI safety failures as a systemic threat, emphasizing loss of control.”