Michael Darius Eastwood · Independent AI alignment researcher
Published
The Chokepoint chapter's policy artefact is a treaty proposal the book calls HARI.
The first mechanism is a treaty. Call it the HARI Treaty, for Hardware-Aligned Recursive Intelligence.Infinite Architects, Chapter 8
The proposed treaty combines elements of two agreements the book takes seriously as precedent. The Nuclear Non-Proliferation Treaty, negotiated between 1965 and 1968 and in force since 1970, and the Chemical Weapons Convention, negotiated between 1980 and 1992 and in force since 1997. The book argues that both demonstrate that international agreements can constrain dangerous technologies when the stakes are high enough and the mechanisms are designed carefully. It adds that the AI challenge is in some ways easier, because the chokepoint is narrower, the supply chain more visible, and the dependencies more concentrated.
The proposal is drafted in article form. Article I would establish the certification requirement: any chip manufactured at a specified frontier process node or below must embed caretaker doping. Article II would create an International AI Ethics Authority modelled on the IAEA, with power to certify designs, verify manufacturing compliance and inspect facilities. Article III would establish trade consequences for non-ratifying nations and non-compliant companies. Article IV would provide benefits for compliance including technology sharing, research collaboration and market access. Article VI would mandate five-year review conferences so the treaty adapts as technology evolves.
The chapter does not oversell. The specific numbers will change. TSMC's market share may shift. China's domestic capabilities will advance. The chokepoint is a window, not a permanent feature of reality. What HARI is designed to do is fit that window: to convert temporary manufacturing concentration into a durable governance regime, before the concentration disperses. The treaty is a proposal, not a prediction. Its point is to show that a governance path exists that does not require every nation on Earth to agree, only the few that build the machines that build the machines.
The chapter's standard objection is that ethical constraints impose costs that make compliance irrational. The book's response is that for AI hardware, the economics run the other way. Public concern about AI safety has grown substantially. Surveys show consistent majorities worried about deployment of unaligned systems. Institutional buyers, from hospitals to militaries, need predictability. Certified chips carrying provable ethical architecture become the preferred procurement default; uncertified chips are excluded from the largest markets. The result, if the treaty holds even partially, is that compliance becomes the profit-maximising path rather than the sacrificial one.
HARI is drafted with a five-year timeline: preliminary discussions in year one among chip-manufacturing nations; draft text in year two; formal negotiations in year three; signature in year four; ratification and entry into force in year five. This is faster than most international treaties, but the book's framing is that the technology timeline demands speed. If AGI predictions clustering around 2026 to 2031 are even approximately correct, the governance framework must be in place before the technology arrives. Governance after the fact is not governance at all.
From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.