In federated learning, data is distributed across multiple devices or nodes, making secure and efficient information transfer a critical challenge. This requires the advancement of complex encryption strategies that can guarantee secure communications when one or more network sensors (nodes) are compromised (e.g., hacked), and when the network topology is not known a priori. In this article, we consider the q-Composite scheme, where a pair of nodes within proximity must share at least q keys to communicate. We introduce a stochastic optimization model for finding optimal key assignments that produce a desired level of communication security in settings where the network topology is unknown in advance. The model enables secure encryption strategies that are resilient against node capture, failures, and network topology changes. We present computational studies to demonstrate the efficacy of the proposed scheme.