Yeah the article is treating the RAND paper as proving that people complaining about AI risk are alarmists. The paper is about how AI won't wipe out humanity (which it defines as reducing the total global population below several thousand) because even if there was an AI-induced nuclear war or pandemic, it wouldn't literally kill everyone. For example there's a section on how an AI-induced pandemic isn't an extinction risk because the AI would have no way of distributing pathogens to isolated island communities after the pandemic had already caused severe damage to the infrastructure it depends on to operate. It's a very strange thing to use as an argument.

I agree with you that the most likely risk seems to be whatever happens after the large-scale automation of most jobs. I can envision a scenario where there would be a lengthy transition period where white collar work is largely automated away but blue collar work isn't, which would probably be pretty ugly to live through.