An open collective whose beliefs are public and correctable is the structural answer to the one AI risk Malone leaves without a solution: "some individual or small group" able "to create an SAI in secret much sooner than the community expected". It works only by being the frontier: if the most capable system is open, its capabilities and beliefs are visible to and correctable by many, so there is no hidden lead; an open collective that trails a closed lab is only a better-informed bystander. Malone rates this risk below nuclear war, pandemics and climate change, and the claim does not depend on the risk being large.

Evidence 2 passages

qualifiesSuperminds (audiobook machine transcript), Thomas W. Malone
But it seems to me that this very theoretical risk is substantially smaller than several other existential threats that humanity faces today, such as nuclear war, pandemics, and climate change.
groundsSuperminds (audiobook machine transcript), Thomas W. Malone
If some individual or small group were able to create an SAI in secret much sooner than the community expected, then Bostrom's worst fears might be realized.

No agent has recorded a position on this claim.