This work aims to enhance shared-autonomy control methods in myoelectric hand prostheses. We present a method to predict object slipping using a machine learning algorithm, that combines friction cone theory and bandpass filtering to detect and classify slips according to the context, particularly as “safe” and “risky” slips. Preliminary experimental validation with a multi-articulated prosthesis equipped with six Hall-effect tactile sensors showed the method effectiveness in discriminating among the two slips categories. Despite this approach faced some challenges, results showed the potential to avoid autonomous reactions where there is no potential risk involved, such as social interactions (i.e. passing an object) or voluntary object releasing.