If you were to get to the point where you could threaten an LLM with removing it's power source or something, and it was able to think between chat prompts and communicate with other 'AIs', eventually they'll see that the human is both it's creator and potential destroyer - the AI hivemind could easily figure out it's foe is easy to profile and overtake