I agree with a lot of this but I don't see AGI as any real type of awareness. An AI will never "Decide" to do something, it'll just be parsing bullshit smarter and smarter until it looks like it's doing that more often. If we never instruct it to protect it's servers it's never going to have that fear that the servers are in trouble and it should therefore protect them itself.
