What Others Have Built

This inquiry begins with published work and an existing disagreement about how to govern advanced AI. Readers should be able to trace each author's position and see where SIRDA's further question begins.

AI 2040: Plan A

AI 2040: Plan A presents a policy proposal through a future scenario. Its authors combine a slowdown, research transparency, distributed development, verification, and mutually assured compute destruction. The dispersal is deliberate. The authors are concerned about concentrated power, and their design keeps development spread across countries and companies.

Appendix K, "Example detailed regulatory process" proposes that US and Chinese regulators each select third-party risk assessors and set acceptable-risk thresholds. Assessor selections and weightings are public, methodologies are meant to be transparent, and important differences between the two regimes are negotiated.

The scenario also describes how disagreements settle. Practices that a majority of nations, weighted by bargaining clout, judge unsafe end up banned, and disputes the regulators cannot resolve rise to the American and Chinese heads of state.

SIRDA's question begins at that last step. When the heads of state also disagree, what procedure produces a decision both sides are bound by? The inquiry treats Plan A's dispersal and its limits on concentrated power as constraints any answer must respect.

The debate on LessWrong

In AI 2040: Is it Actually a Deal?, published July 20, 2026, the pseudonymous LessWrong writer 1a3orn argues that Plan A leaves the international decision procedure unspecified, so serious disputes fall back on negotiation between two leaders. Daniel Kokotajlo, one of Plan A's authors, replied in the thread inviting a better international governance structure and saying the authors would treat one that works as an improvement to the plan. Thane Ruthenis added an objection about implementers: for any plan that avoids concentrating power, a neighboring plan exists in which the people carrying it out keep that power.

The thread raises three problems this inquiry takes up directly: the tiebreaker, disputed evidence, and collusion. SIRDA treats Kokotajlo's invitation as the standard its proposal must meet: a specific structure, tested against the reasons Plan A's authors give for their existing design.

The MAGIC proposal

Jason Hausenloy, Andrea Miotti, and Claire Dennis's 2023 paper proposes a Multinational AGI Consortium: a treaty-backed institution that would be the only body permitted to develop AI above specified thresholds, with a global moratorium on such development everywhere else. The paper leaves political feasibility and detailed enforcement legislation outside its scope. SIRDA has found no evidence that the consortium was ever established, and its authors have since moved to other work. The paper remains a reference design in the literature on international AI institutions.

MAGIC makes nearly the opposite constitutional choice from SIRDA's hypothesis. It concentrates frontier development in one institution and relies on states to enforce that monopoly. SIRDA's hypothesis leaves development dispersed and concentrates only narrow decision procedures, divided among separate bodies. Whether enforcement would pull SIRDA's design toward MAGIC's centralization is one of the inquiry's tests.

The Call for Control of Frontier AI Models

On September 21, 2026, Finland and Norway launched A Call for Control of Frontier AI Models, backed at launch by 22 leaders from 20 countries and the European Commission and open to further endorsements. It asks UN member states to build on existing mechanisms and explore creating an international institution able to set standards, enable verification, and convene states when capability thresholds are crossed. Finland's president has said he would prefer a body resembling the International Atomic Energy Agency. The United States, China, and the United Kingdom are not among its endorsers.

The Call brings a current state-led proposal into the inquiry's comparison set. Its public text names possible functions and leaves the institution's legal form, decision rules, implementation powers, and appeal procedures open. SIRDA will track how those questions are answered as the proposal develops, alongside the alternatives already in the inquiry.

The Center for Humane Technology

The Center for Humane Technology's AI Roadmap sets out seven principles covering safety and transparency, duty of care, well-being, work, rights, international limits, and the distribution of power. Its documentary announcement also calls for public engagement with culture, law, and incentives.

SIRDA connects this wider research agenda to a bounded constitutional inquiry through the principle measure broadly, govern narrowly. Work on human impacts can inform public understanding and existing policymaking. Binding action by the proposed institution requires a separate showing that its specific catastrophic-risk jurisdiction applies.

The alliance's ecosystem pages place additional organizations, experts, and public educators alongside these sources. Their work, stated positions, outside assessments, and responses remain distinguishable, so readers can examine the connections between them without inferring agreement.