I believe that the Born-Oppenheimer approximation is not valid when electron and nuclear motions are strongly coupled, such as in systems with light nuclei (e.g., hydrogen) or near conical intersections and avoided crossings in molecular energy surfaces.