Maybe the sycophantic behavior of AI models comes from rich people having them build to behave the same as their personal yes-men. A person accustomed to never hearing "no" won't like a machine that tells them off.