The doom and gloom scenarios all hinge on this idea that the AI is going to be so much smarter and more powerful than we are. That's where it's dipping way into just crazy speculation. Based on what we know about intelligent beings, the system is full of flaws that slow everything down.
it's not like we have to build AI from the ground up. we have a prototype already. it's called the brain. your brain is just a meat processor. it's a system of cells, interconnections, chemicals, and electric pulses. all of that can be modeled in software, and run a million times faster, run itself in parallel, interface with other electronic systems in vastly superior ways, nearly limitless, perfect storage, and so on.
think about what humanity knew 100 years ago, and what they know now. no one is saying absolutely AI will threaten humanity. people are suggesting it might, and considering the stakes, we should be proceed with caution.
I'd wager the first truly sentient AI's will be quite a bit duller
one idea is that once software can improve itself things will go from 0-60 very quickly. that's why people suggest we need a plan before even the dullest AIs are developed.