This PhD thesis – composed of a series of five papers tied together by an introduction and conclusions – explores the functioning of an Incident Reporting System used in a high-risk organization. It aims at understanding how safety is actually constructed by and through this reporting system and what it actually does to the meaning of safety when it prescribes a set of sociotechnical practices. From a conceptual point of view, this PhD thesis departs from classical risk or vulnerability analyses which tend to measure factors in order to produce what are supposed to be “objective” results and “manageable” recommendations for decision makers. By relying on a constructivist understanding of vulnerability, this research explores and characterizes situations and events the consequences of which are ambiguous and the probabilities of which are uncertain. It thus broadens the scope of classical risk or vulnerability analyses and complements it with a qualitative and discourse-based attention to organizational, cultural and ethical questions.