To overcome the vital shortcomings of traditional BDI Agent in the application of VO, we proposed a Policy-Oriented Autonomic, Controllable Agent Model. To enhance the controllability of the Agent, a policy was used to restrain the formation of Agent motivation and to affect the formation of their intentions and behaviors. This model used a hierarchical policy structure, with different policies representing the interests of different partners in the VO. In order to solve the conflict between individual interests and organizational objectives, the policies were divided into two different layers of organizational policy and individual policy. By using extended Defeasible Logical framework, we established a formal theory for the model with linear time computational complexity. This model maintained a certain degree of autonomy and at the same time enhanced the controllability at the organizational level.