Contemporary studies of genocide have found military capabilities to be inconsistent predictors of state-sponsored killings. We suggest that these empirical inconsistencies stem from the fact that government strength can serve two opposing purposes. Some level of armed capabilities is necessary for a state to remain viable and to provide internal and external security. Yet armed government personnel can be deployed to repress and destroy segments of the public. We identify conditions under which an executive is more likely to use security forces for private-interest killing rather than public protection. We hypothesize that unconstrained leaders are more likely to use their putative security forces to initiate genocide and remain in power. An analysis of state failures that lead to genocide robustly supports the idea that the effect of increased security forces on the risk of genocide is conditional on institutional executive constraints.