The main issue is that if a goal of a general artificial intelligence is not aligned with the best interest of humans.

Think about ants and humans. If we want to build a road, we don't really care about the ants that were living there.

An AI can affect the real world through the internet. Imagine all the services available now, and that you can pay with Bitcoin, plus emails, plus social engineering, etc. All the scams in the world are nothing compared to what an AI can do.