The use of power assistive devices that use surface electromyography (SEMG) signals may be limited by the noisy nature of SEMG signals. The aim of this study was to investigate the variation in human movement stability while the amount of SEMG-based assistive power was changed. A robotic device provided a torque that was proportional to the torque estimated by SEMG for assisting human movements, and 12 volunteers participated in the elbow flexion experiments. The maximum finite-time Lyapunov exponent (MFTLE), the average logarithmic rate of the divergence of neighboring trajectories, and the variability of the kinematic data were used to quantify the stability of the assisted elbow movements. The stability provided by the MFTLE decreased as the amount of assistive torque increased with respect to the amount of human torque. The kinematic variability increased with the increase in assistive torque. Therefore, by ensuring that the amount of SEMG-based assistive torque is less than the amount of human torque, the assistance may provide relatively natural movements. This study is the first to quantify movement stability as SEMG-based assistive power is applied. This study can provide a foundation for determining the appropriate amount of SEMG-based assistive power.