I'm finding the logic opposed to a mandatory kill switch slightly strange.

I appreciate a belief it won't help adversarial use of AI outside your control, but in what way does it either hinder, or not help malbehaved AI in your direct control?

This is a trend I see increasingly online.

  A: "let's do a good big thing"

  B: "but it has this 2nd, 3rd order possible hypothetical downside and anyway, as a doomer, it won't work"

It's as if ten year old bored kids are steering the bus.

  Omnes: we didn't do the good big thing.

Except they don't stop the bus falling off a cliff.