Yea. The problem is never with the model - it is with the people. In a way, models trained on huge corpus of data is the most "democratic" way of representing reality - removing "biases" from it is castrating it. Those that are to exploit those biases needs to be dealt with on an individual basis.