Remote disablement and control of autonomous cyber-physical systems is possible through the external manipulation of sensory subsystems. Many modern autonomous systems utilize neural networks to fuse and parse data from sensor input streams. We suggest that the application of probabilistic neural network models increases the robustness of machine learning in sensory subsystems. This study compares Probabilistic Backpropagation (PBP) and equivalently sized non-probabilistic models at processing datasets injected with normally distributed noise. Our results suggest that PBP performs with a smaller RMSE and that its estimate of the posterior uncertainty of weights provides insight to the trustworthiness of the model.