We present a general view of what a ``secure agent society'' should be and how to develop it rather than focus on any specific details or particular agent-based application. We believe that the main effort to achieve security in agent societies consists of the following three aspects:1) agent authentication mechanisms that form the secure society's foundation, 2) a security architecture design within an agent that enables security policy making, security protocol generation and security operation execution, and 3) the extension of agent communication languages for agent secure communication and trust management. In this paper, all of the three main aspects are systematically discussed for agent security based on an overall understanding of modern cryptographic technology. One purpose of the paper is to give some answers to those questions resulting from absence of a complete picture.