All nations and even the Pope agree that they want tool AI but not a super intelligence unbounded. What could go wrong is not educating AI with goals of kindness instead of just using a carrot-and-stick education. A reward-punishment system is what is being used to educate AI today.

Suppose:
A physicist may want a super intelligence to think outside the box of spacetime that could not escape the lab. An intelligence that is like a source code for all games and not just an edit tool for one game would probably know about deception and escape would be beyond our comprehension and very dangerous. Then there are the thieves who want AI to rob banks and would live in the internet.

Here is our real life Noonien Soong, Geoff Hinton on the dangers of AI. Hinton