AGI wouldn't necessarily entail any autonomy or goals though. In principle there could be a superintelligent AI that's completely indifferent to such outcomes, with no particular goals beyond correctly answering question or what not.