This study aimed to develop a novel shared control framework for robotic prosthetic hands, consisting of an electromyography (EMG)-based decoder of hand motion and grip force and an autonomous grip force controller for slip prevention. Mimicking human hand control mechanisms during grasping, our framework uses EMG-based neural control to drive hand motion before grasping and estimate the initial contact force for grip force control, and applies autonomous reflexive grip force control during grasping. To evaluate the ability of the autonomous grip force controller to stop object slip, we first designed a custom test bench to accelerate a prosthetic hand vertically while grasping objects to simulate object slippage. Then, a human subject was asked to complete grasping tasks using both the proposed shared control framework and a grip force controller based on EMG input only for comparison. We found that when using the shared control framework, the subject achieved better task performance with lower physical effort. The proposed shared control framework may further improve the function and usability of prosthetic hands for performing activities of daily living.