The real question may not be whether AI serves society or shareholders, but whether we are designing clear execution boundaries that make responsibility explicit regardless of who owns the system.