My understanding is the 1/r^2 behavior comes from solving for Green's function of a PDE. Gravity is similar to Gauss's law (i.e. 1/r^2 behavior of charged particles) in the underlying PDE. So something in that equation would have to change.

Apparently it makes very accurate predictions though, so it's not just a matter of ad hoc tweaking some law to fit reality... It does seem to capture some regularity/mechanics of nature.