With the increasing development of self-driving vehicles, there are various substantial risks in the interaction between automated driving technology and conventional transport system along with users, how to prioritize the risks involved in self-driving vehicles is regarded as a considerably complex multi-criteria decision making (MCDM) problem. In response to this, this study aims to propose a novel hybrid MCDM method for quantitatively identifying and prioritizing major types of risks related to self-driving vehicles, and addressing the main issues in the decision process, including information loss, criteria attribute and risk preference. The interval type-2 hesitant fuzzy linguistic term set is used to express double uncertainties on the mutual influence degree among criteria performances associated with each alternative. A combination weight method is developed to measure criteria weight and to investigate the interaction between criteria. And a cumulative prospect theory modified PROMETHEE II model considering criteria properties and risk preference simultaneously is developed to prioritize risks in self-driving vehicles. The findings of this study indicate that the Cyber Attack Risk (A2), Reputational Risk (A1) and Internet Outage Risk (A3) are specified as the top three prioritized risks. The sensitivity analysis illustrates that the final prioritization of risks is influenced by changing criteria weight, criteria properties and risk preference. The comparative analysis demonstrates that the proposed model turns out to be very practicable and feasible, due to its large distinction degree.