Trustworthiness Preservation by Copies of Machine Learning Systems