Censoring models is not safety but safetizm. It is the TSA of the AI world. Safety is making sure the model cannot do anything not allowed even if it wants to.