ABSTRACT Deploying defense strategies in dispersed computing consumes limited resources, a challenge exacerbated when nodes have only partial information about the system state. This article proposes a security defense strategy for dispersed computing based on Backward Mean Field Games. We first analyze the problem of limited node perception ability during attacks, and construct a backward stochastic differential equation to model the evolution of node resource states. We define an individual cost function intended to optimize resource consumption for the defense strategy. Then we find the optimal decentralized defense strategy and prove that it is ‐Nash equilibrium in the limit system. Finally, simulation experiments validate the strategy's effectiveness, demonstrating that the average system state rapidly stabilizes, and individual nodes robustly track this mean‐field trajectory toward their security targets, even under local attacks. These findings confirm that the strategy effectively guides nodes toward their terminal security targets, highlighting the strategy's good performance and robustness in balancing security needs against resource consumption.