If you start to let AI decide or influence policy, it can kill people. This seems very simple to me...

1. Put AI in charge of health policy 2. Give it a naive misguided charter/goal eg eliminate suffering 3. Euthanasia for everyone

This is simply AI optimizing for the goal - the goal is the evil/naive/misguided part, not the AI.

Those who aim to "eliminate suffering" are the issue here to be clear, but unfortunately they are a growing group.