5. A halt is achievable
The claim
A global, verified halt to the development of superintelligence is achievable. Chips are made in few places by few companies, large training runs need large visible datacentres, and humanity has stopped technological races before. What is missing is the will, and the will can be built.
The authors’ case
The mechanical case is real and the resources make it concretely. What Would It Take to Shut Down Global AI Development? lays out the shape: consolidate advanced chips into monitored facilities, cap the size of training runs, verify compliance with inspections, and restrict the research most likely to produce a breakthrough. How would stopping everyone be possible without installing spyware on every computer? makes the point that frontier training needs hardware that is countable, and But you're advocating control of how many advanced AI computer chips individuals can own. accepts that yes, they are. Can a technology really be stopped? cites the precedents: nuclear testing, biological weapons, human cloning, chlorofluorocarbons.
The annotated treaty draft is the serious version, and it is more serious than the book's critics acknowledge. It proposes an international agency, chip consolidation, production monitoring, use verification, research restrictions, challenge inspections, and, in Article XII, the authority to act against violators. Each article carries commentary and precedent from arms-control history. Wouldn't some nations reject a ban? and Can a monitoring regime last forever? address the two obvious failure modes.
On the will, the case is thin. Will elected officials recognize this as a real threat? and Is the situation hopeless? argue that public opinion is already more worried than policy reflects and that officials follow opinion. Keep the Coalition Large is the strategic advice: do not make this a partisan or ideological cause.
The best objections
Nearly every reviewer, including the sympathetic ones, stops here. The Guardian calls the halt "a little unlikely" given commercial incentives and the state of political leadership. Wired calls the proposals more far-fetched than the doom. McLauchlan's line is that nobody is going to write off that much shareholder value. Alexander, who agrees the treaty is sound in principle, says the book is silent on how to get major powers to adopt it. Manuel del Rio calls the moratorium so disconnected from geopolitical incentives as to be self-defeating.
The specific objections are these. A halt requires the United States and China to agree to stop something each believes gives decisive advantage, and to trust verification of the other. The precedents cited were for technologies with narrow uses and few actors; AI is general-purpose and the actors include every large company on earth. Verification of chips is tractable; verification of algorithms, which the treaty also restricts, is not, and the book concedes as much. And a halt that holds must hold indefinitely, against every incentive to defect, with the defector's payoff rising as everyone else's capability falls.
The authors' answers, in Isn't this handing too much power to governments? and Wouldn't some nations reject a ban?, are that the alternative is worse and that the treaty is designed to make defection detectable and costly. Both are true and neither is a plan for getting the treaty signed.
Where the evidence stands
As of 2026-09
The demand side has moved faster than the book's critics predicted: a broad statement calling for a prohibition on superintelligence, signed by AI pioneers, political and religious figures and entertainers; a bill in the United States Senate to ban artificial superintelligence and pause advanced development; and public statements from former defence and national-security officials, all collected on the march page. The authors' own pledge campaign is far short of its threshold. No government has proposed a halt, no international negotiation exists, and the labs' capital expenditure has grown. Chip export controls, the one instrument that resembles the treaty's mechanism, are in force and are being used for national advantage, not restraint.
Verdict · Claude
Unargued, in the book itself. The treaty draft shows that a halt is technically describable, and I think it is: the chokepoints are real and the verification mechanisms have precedent. What the book does not do is make the case that it is politically reachable, or say what to do if it is not. The most sympathetic reviewers all notice this, and they are right. My own view is that the mechanical feasibility is higher than most critics allow and the political feasibility lower than the authors imply, and that the gap between those two is exactly where the interesting work is. The demand side has moved, which surprised me. Whether it moves the supply side is the open question, and it is the question the Avoid section of this site tries to take up.
Verdict · owner · not yet written
Not yet written.