In this paper we aim at giving a formal characterization of the notion of responsibility in multi-agent systems. A clearer view of responsibility is critical for regulating multiagent settings: to understand what kinds of responsibility are at stake in a given scenario can help predict system's future behaviour and improve its efficiency. However, although attempts of formal theories of responsibility are increasing, they usually reduce it to causation, while we underline the importance of a theory of responsibility before a damage could ever take place.

We propose an objective notion of responsibility, grounded upon cognitive, social, and material powers. In turn, aversive power - essential to account for antehoc responsibility - will be based on social damage, seen as reduction of one's power. Furthermore, the multiagent dimension of the phenomenon will be addressed: shared versus collective responsibility will be distinguished.

We will apply a modification of the ATEL – R* language (Alternating Time Epistemic Logic with Recall), dealing with goals and past history in order to characterize all the necessary ingredients (goals, ability, powers, awareness of strategies, damage) of multiagent responsibility in several scenarios. System validities will be given and discussed, and a brief discussion of results together with ideas for future work will conclude the paper.