It seems that during the displacements for the numerical MP2 hessian the point group changes from C2v to Cs (or C2) and then the SCF restart from the previous orbitals fails with above error message.
The symmetry is autodetected.
An immediate remedy is a) forcing C1 symmetry at an increased computation cost. (put symmetry c1 in molecule block) or b) avoid calculating the Hessian
I believe @jmisiewicz is an expert on the FINDIF issue.