Animals encounter several dangerous situations during their lives, in response to which appropriate defensive responses are essential for survival. The selection of a proper defensive response to cope with each dangerous encounter relies not only on the recognition of specific threat characteristics and its context, but also on the detection of features shared by different threats. With a novel paradigm that exposes mice to multiple threatening situations, we recently demonstrated that neuronal populations in the dorsomedial prefrontal cortex (dmPFC) encode both a general danger state and specific threats. However, how these general and specific danger representations arise from the combined activity of different neuronal populations and guide the selection of adaptive defensive responses remains to be elucidated. To investigate this, calcium imaging on pyramidal (CaMKII+), parvalbumin- (PV+) and somatostatin-expressing (SST+) neurons of the dmFPC was combined with the novel behavioral paradigm. Although the presence of a threat could be predicted from the patterns of population activity of all neuronal types, threat-related information was found to be more robustly encoded by the SST+ population than by the PV+ and CaMKII+ ones. Overall, our results indicate that the population of SST+ neurons generates specific representations while those arising from PV+ neurons are mainly unspecific. This data suggests the presence of distinct information in different dmPFC neurons allowing a collective encoding of both general and specific danger representations, which contributes to the selection of adaptive defensive behaviors.