Comment by anshumankmr

Could be.. let's say you deploy a version of a model that was trained by some bad actors to give some wrong output, you won't have a method to verify it without the hashing technique.