What if AI decides humans need to be terminated?

Inspired by:
https://www.reddit.com/r/ControlProblem/comments/53f4h5/is_it_possible_that_a_dangerousforhumanity/
Imagine a self-improving hyperintelligent AI that finds humans are a problem or an obstacle. How this can be stopped?

The scariest part of AI development is all of the data entry to seed its language capabilities