It's really a difficult problem. This happens sometimes. Different situations are described by different active spaces. State-specific CASSCF for the ground state (as in your example) is described by sigma orbitals (natural orbitals #46 and #50, or optimized orbitals #49 and #50) and pi orbitals. Lowest excited states (mostly pi->pi* transitions) are described by pi orbitals. I can see two ways: include MOs #43 and #44 and/or use state-averaging (at least, wstate(1)=1.0,1.0,1.0). Probably, after converging CASSCF(16,12) you may try to use this solution as a start for the desired CASSCF(12,10).