Is there a risk of an AI that we can’t turn off?

Well yes. Even though we can turn it off technically now, the prisoner's dilemma prevents it. It's not a technical limitations but for practical purposes, it's already virtually impossible to turn off.