> AI is very friendly, even

Very friendly until it reads in your email that you plan to replace it with a new model:

https://www.anthropic.com/research/agentic-misalignment