Using the model to find the heavy atom sites and then calculating phases on those sites (without model) might help. ( I believe phenix will output a phase combined map (model + experimental phases)).
Correct, this is the default behavior if you provide a model directly to AutoSol (or Phaser). There is however a semi-automatic way to use the model phases to find sites without biasing the experimental map; see the documentation here:
(The AutoBuild step is optional - you could just look at the density-modified map after AutoSol is done.)
-Nat