An important problem faced by children is discriminating between entities capable of goal-directed action, i.e. intentional agents, and non-agents. In the case of discriminating between living and dead animals, including humans, this problem is particularly difficult, because of the large number of perceptual cues that living and dead animals share. However, there are potential costs of failing to discriminate between living and dead animals, including unnecessary vigilance and lost opportunities from failing to realize that an animal, such as an animal killed for food, is dead. This might have led to the evolution of mechanisms specifically for distinguishing between living and dead animals in terms of their ability to act. Here we test this hypothesis by examining patterns of inferences about sleeping and dead organisms by Shuar and German children between 3 and 5-years old. The results show that by age 4, causal cues to death block agency attributions to animals and people, whereas cues to sleep do not. The developmental trajectory of this pattern of inferences is identical across cultures, consistent with the hypothesis of a living/dead discrimination mechanism as a reliably developing part of core cognitive architecture.