You are assuming that superintelligent AI will follow every command of users outside the labs.