That's the alignment problem. We intended a certain set of rules but didn't define them completely, or there were conditions we didn't consider. An AI wouldn't

That's the alignment problem. We intended a certain set of rules but didn't define them completely, or there were conditions we didn't consider. An AI wouldn't have to maliciously break a rule to go wrong. The point is that the system could so exactly what it was supposed to do, it plays within the given rules but the outcomes aren't what we wanted or intended.