A fleet of UAVs flying according a planned mission in hostile environments must optimize its configuration, so that the UAVs auto-protection systems ensure the fleet safety. Such an optimization can be done in the mission planning phase, and must also be reactively updated to tackle unpredicted threats. In this paper, we present an approach to the problem of selecting autonomously the fleet configuration. Algorithms that explicitly consider models of the threats and of the countermeasures systems are presented, and integrated within a global decisional architecture that ensures reactiveness and constraints satisfactions.