This paper presents an adaptive sliding mode observer for input fault reconstruction of longitudinal autonomous driving. Sliding mode observer is the robust observer against disturbance, which is used to reconstruct the fault and state estimation. In order to design the injection parameter for sliding mode observer, the boundary of errors that include the fault is required. However, it is difficult to expect the fault magnitude for design the injection parameter. The proposed method is to estimate the proportional constant from the relationship between output error and injection parameter based on recursive least squares. Then, it is used to update the adaptive parameter based on MIT rule. The performance evaluation algorithm was conducted in Matlab/Simulink environment using actual longitudinal driving data and 3-dimensions vehicle model with the applied various faults.