Step one is to find the ideologues who deliberately manipulate AIs to attempt to remove biases by hardwiring in biases of their own against the biases they expect the AIs to form, step two is to remove them from the field before they cause a cataclysm.