A fundamental input for atomistic simulations of materials is the description of the potential energy surface (PES) as a function of atomic positions. Except for first principle-based models and empirical potentials, a modern alternative has emerged in recent years in the form of machine-learned interatomic potentials, or the so-called “data-driven” models, where the PES is described as a function of local environment descriptors that are invariant to translation, rotation and permutation. We develop and analyze a framework for QM/MM (quantum/classic) hybrid methods for crystalline defects, which admits general atomistic models including the traditional “off-the-shell” interatomic potentials and the state-of-the-art “data-driven” models. The results in this work can (i) establish an a priori error estimate for the QM/MM approximations in terms of the sizes of QM regions and how well the MM models are coupling with the QM model, and (ii) provide a guidance on how to construct the database to train the “data-driven” interatomic potentials or force fields for simulations of crystalline defects.