Comment by aesthesia

If things were implemented that way, you would see vastly more complaints about overeager refusals. For at least some deployments, the safety classifiers are based on internal LLM activations, so the input does get processed by the model.