During disasters, people, confronted by life-threatening situations, deal with high levels of stress. While some groups of individuals will be able to cope and adopt thoughtful responses, others will experience acute stress reactions that will disrupt their psychological and physiological functioning leading to irrational responses such as terror, agitation or panic flight (Chidiac & Crocq, 2010). These last reactions are difficult to manage for the emergency services. They often require early assistance provided by medico-psychological emergency cells. Several factors can influence people’s reactions in face with a disaster. Beyond the intensity, the frequency, the temporality and the spatiality of the event (Dubos-Paillard et al, 2021), some factors such as education, risk culture, preparedness can influence the human toll of the disaster. This paper proposes a theoretical reflection on two factors: the diffusion of information via a warning system reaching a large number of individuals (Douvinet, 2020) and a more localized but on-site intervention by the police and rescue services. We assume that they can influencepositively the dynamics of behavior during the event, but that their effectiveness depends on the period in which they are implemented. We base our reflection on a dynamic multi-compartments model, the Alert-Panic-Control (APC) model, which allows to model the diversity of human behaviors and their evolution during a disaster (Provitolo et al, 2015; Lanza et al, 2020; Verdiere et al, 2021). Inspired by epidemiological models, it makes it possible to simulate different scenarios in which the population can evolve between three kinds of behaviors named Alert (awareness of the danger), Control (sheltering, evacuation, etc.) and Panic (disordered agitation, panic flight, etc.). The objective is therefore to estimate the best moments to implement these two solutions, considering uneven levels of efficiency in order to limit potential “panic” behaviors.