In post-processing this would be relatively easy – just “turn off” the interactions of atoms C (set coefficients and charges to zero or use neigh_modify exclude or something else) and rerun the trajectory, and you will have per-atom forces that are purely A-B interactions.
During the simulation itself? Doesn’t seem possible. But you could use compute coord/atom to calculate the average number of B atoms around each A atom and their average distances, and even try to chunk that, and then work out the average forces as a function of coordination-density.