A real-time online safety supervisory-control strategy based on fuzzy logic is presented to improve the safety and stability for the upper-limb rehabilitation robot in clinic application. During the robot-aided impaired limb rehabilitation exercise, the impaired limb condition impacts the control performance. An intelligent safety supervisory fuzzy controller (SSFC) is designed to enhance the movement stability and safety in emergent condition. Firstly, the movement features are extracted to evaluate the stability of the impaired limb, subsequently the proposed safety supervisory fuzzy controller intelligently adapts the desired control force to a reasonable disturbance or responds to a sudden event in time. Secondly, a position-based impedance control strategy is adopted to achieve the compliance between the impaired limb and the robotic end-effector. Experimental results show the effectiveness of the proposed method for achieving the safety and stability of the rehabilitation robot.