I more meant, if a model is only trained on 100% absolutely positive data concerning a specific subject, including any data that implies certain taboos it must

I more meant, if a model is only trained on 100% absolutely positive data concerning a specific subject, including any data that implies certain taboos it must avoid at all costs, the model will embed that strong bias in its learned parameters.