Model behaviour is determined by your dataset, nothing else

[navjack27]

Literally no duh. That's why it's impossible to totally correct model "lies" or image generation censoring after the fact. You need to retrain on better data.