The proliferation of AI in Machine Learning as a Service (MLaaS) platforms has brought increased attention to the security and privacy of AI models, particularly within mobile edge computing environments. The ability to easily harness the analytical power of machine learning highlights privacy concerns when models are trained using sensitive data, as recent research has shown that models may retain and subsequently leak this information. Exposure of a participant’s membership in the training set of a sensitive model, known as a Membership Inference Attack (MIA), can have severe personal and legal consequences. In this work, we explore using an empirical pruning approach to increase model robustness against privacy attacks like MIA, while simultaneously enhancing model efficiency-a crucial aspect for mobile edge and IoT deployments. Using our approach, we observe a moderate decrease in MIA recall with no comparable changes in MIA precision. With further refinements to our approach, we believe that model pruning can be a useful tool for protecting models against privacy attacks, as well as contributing to the security design of MLaaS applications in IoT and edge computing environments.