Even when the user mentioned specific political figures and recent attacks, the AI continued to make dangerous and hateful ...