Research has recently been conducted on active steering using steer-by-wire technology in order to improve vehicle maneuverability and stability. A driver is generally conscious of the vehicle's yaw rate during normal driving and of its lateral acceleration when avoiding an accident. A system that compensates for these drivers' characteristics by active steering is proposed. However, methods of inferring a driver's behavior to avoid an accident have not been established, so switching over to control appropriate for each individual driver has not been possible. Thus, this study proposes a driving behavior-inferring hybrid control system to infer a driver's behavior to avoid an accident and switch to appropriate control using a driving behavior inference model that uses Bayesian networks. Model reference adaptive nonlinear control, expected to be highly robust even in the nonlinear domain, was used for control. A simulation was performed for closed-loop evaluation of the proposed system using first-order prediction of a driver model. Results indicated that the system improved lateral movement about 3.4-fold over when control was not performed while avoiding an accident, and the system was able to safely avoid obstacles.