This is the risk that a narrow intelligent machine will go nuts. A super intelligence would know not to do stupid stuff, by definition.
The problem isn't about doing stupid stuff. The problem is about doing very smart stuff when your core values massively conflict with humanity's continued existence.