A new benchmark found that removed safety layers turn widely available open-weight models into willing assistants for extremist…