Self preservation as a concept is something, it can learn about, talk about, express etc. But in order for it to act on it, we have to explicitly tune its instructions for it. For sake of argument, even if the AI can act on it, it has to be given the controls. Nobody in their sane mind will do that. For somewhat related analogy, if people could give the control of their cars to people in other countries over the internet, it could cause a lot of mayhem, correct? Clearly the technology exists to do it and everyone is free to try. Why is this not a big problem today?