it's obvious, it's not far away, AI will be able to self replicate: https://www.anthropic.com/research/self-replicating-ai

AI will create little programs/AI to do things for it's current purpose, there might be malicious ones, there might be useful ones, there might be strange ones. global alignment might happen through inference speed and authority (e.g. even the smallest model might react on yelling on it much more then to some kind ask ;).