Limitations on Safe, Trusted, Artificial General Intelligence