> I don't even think it'll be from terminators and nuclear wars and that sort of thing
I do. And I don't even think the issue is a hostile AI. There are 8 billion people in the world. Millions of those people have severe mental issues and would destroy the world if they could. It seems highly likely to me that AI will eventually give at least one of those people the means.
This is explored in the Vulnerable world Hypthesis-argument by Nick Bostrom [1][2]
[1] https://en.wikipedia.org/wiki/Vulnerable_world_hypothesis
[2] https://doi.org/10.1111%2F1758-5899.12718