The "normalization of deviance" is a direct result of the "Loss of God" of secular society. There is no way to have one without the other...its baked into the cake.

In a very very analogous way the "safety of LLMs" is a mirror of this...A LLM is just statistical token generation machine, its going to generate output based upon its training data. If you want a LLM to be "realistic" then you need to train it on realistic human conversation, which will include unsafe thoughts. Its unfortunately again, baked into the cake.