More RLHF, More Trust? On The Impact of Human Preference Alignment On Language Model Trustworthiness