Analytic

The remedy in If Anyone Builds It, Everyone Dies is itself a collective, public verification regime, not a concentrated one: all compute that could train stronger AI "gets consolidated in places where it can be monitored by observers from" multiple treaty powers, holdouts get "an offer to join the treaty on equal terms", and enforcement rests on public justification ("Clear communication is key."). What the authors reject is collective steering of a superintelligence ("even an international committee would have no hope of shaping a super intelligence"; "It doesn't matter who's in charge"), not collective oversight. So the dispute between the book and collective, correctable AI is not concentrated versus distributed control. Both put a shared, multi-party verification layer at the centre. The dispute is what that layer verifies: that nothing more capable is built, or that what is built stays correctable. The book's case is that only the first is possible with current knowledge; ours has to show the second is, or concede that our layer's first job is verifying a halt.

Alignment dynamics