Studies on the phenomenon of anthropomorphism underpin the strategies by which social robots are designed. This allows for greater acceptance and trust in the machine, but may lead to manipulative drifts and to the distortion of the perception of reality. This paper argues that anthropomorphism cannot be fully controlled and surely not eliminated, because it represents a characteristic and irreducible feature of humanity. In this sense, it represents the materialisation in the context of human robot interaction (henceforth HRI) of the more general concept of human vulnerability. Nevertheless, it is affirmed the necessity to evaluate how and to what extent the design features implemented in different robots may elicit, target, support such a phenomenon (and thus human vulnerability), and how they can be handled for the protection of the fundamental rights of individuals, first of all dignity and psychological integrity. To this end, a Vulnerability-oriented Impact Assessment is suggested.