Interatomic potentials are surrogate models that govern the physics of atomistic simulations and provide immense computational savings compared to ab initio molecular dynamics (MD). However, these reduced computational requirements may come with noticeable reductions in accuracy over ab-initio methods. Recently, there have been large advances in machine learning techniques, which have steadily been adopted by the material science community, ushering in the data-driven field of material informatics. Our goal is to develop a machine-learned interatomic potential (MLIP) for the silicon carbide (SiC) system, intended for simulating extreme environments (e.g., shock loading). We use an artificial neural network trained on ab initio calculations to fit an MLIP for the SiC system. The training data includes multiple polytypes as we aim to capture phase transformations observed under extreme loading conditions approaching ab initio accuracy with lower computational costs compared to current potentials. While MLIPs may not be as physically interpretable as their traditional counterparts, we expect that their flexible form will enable the identification of connections across length scales that are otherwise missed in more traditional approaches. Efforts for SiC MLIP development will be discussed, along with network architecture optimization.