Modeling damage initiation and progression in continuous fiber reinforced ceramic matrix composites (CMCs) needs to be improved before CMCs can be widely adopted in critical aerospace applications. While damage mechanisms in CMCs are heavily dependent on the microstructure, we develop a continuum damage model that homogenizes governing microstructural mechanisms for application at the engineering scale. The continuum damage model operates on the ply level, under 2D plane stress assumptions, derived using a thermodynamics-based approach. To facilitate the development of the unknown damage portion of the model, we employ genetic programming based symbolic regression (GPSR) to learn the unknown damage model tensors from SiC testing data. Known elastic material behavior is separated such that GPSR specifically learns the corresponding damage tensor. In addition, training constraints using physical knowledge are enforced to promote the evolution of models that are physically valid and interpretable. Using this framework, GPSR learns damage tensors that include ply level damage behavior and interactions between off-angle plies. The resulting damage tensors provide physical insight into damage mechanisms and interactions that were missing from previous models.