The design of automated online security and intrusion tolerance solutions for computer systems and networks is a theoretically achievable objective. However, after two decades of research, this objective has still not yet successfully passed the practicality test to become widely adopted. We believe that the main reason behind this gap lies in the misleading definition of autonomous security that precludes human intervention in trying to achieve unrealistic level of full automation. In this position paper, we discuss a novel research direction to address this challenge through a coherent integration of human intelligence. In particular, we identify the key questions to be answered in order to enable human intelligence to be leveraged in a seamless manner and without requiring explicit heavy human involvements. We illustrate our discussion through the preliminary design of a semi-automated intrusion detection and response system that incorporates reasonable human input to achieve higher levels of intrusion tolerance efficiency.