I'm not so sure that the argument that some humans will survive would placate anyone who is worried about a catastrophic outcome from AI development. You could make the same argument for climate collapse or global nuclear war, but I also don't think that's gonna make anyone feel better about those scenarios

My most likely doom scenario is that a frontier model convinces the CIC to press the nuclear retaliation button. Maybe this will only kill 90% of humans, but as far as I am concerned, that’s good enough.

I think a realistic scenario is mass murders triggered by AI psychosis. Some weirdos start an AI cult and end up killing a bunch of randos because the AI said doing so would break the matrix or something similar. I am actually surprised this hasn't happened yet.