The Founding Inquiry
Should Humanity Create a Government Without Territory?
What powers must exist to govern catastrophic AI risk, who should hold each one, and what should constrain it?
Advanced AI concentrates risk in a small number of increasingly capable systems while the means of governing that risk stay scattered. Labs build the systems. Independent evaluators can identify particular dangers. Compute and infrastructure providers control important operational chokepoints. Governments hold legal authority within their borders. The people exposed to the consequences hold a legitimacy that none of these actors can supply for themselves.
Transparency shows what participants are doing. Verification establishes facts. Standards set common expectations. The hardest problem begins after all three, when informed and legitimate participants read the same evidence and disagree about what it requires, and delay or unilateral action could be catastrophic.
SIRDA tests one answer to that problem: a government defined by capability rather than territory. Its jurisdiction would cover only catastrophic and existential risk from advanced AI, and its powers would be divided among separate institutions for evidence, threshold setting, adjudication, enforcement, appeal, and oversight. Everything outside that jurisdiction would remain with existing governments, institutions, companies, and people.
The Hypothesis Must Earn Its Complexity
Less ambitious arrangements already exist or could be built: an international standards regime in the style of the Basel Committee, stronger national regulation, a treaty institution, and voluntary coordination among developers. Each serves as a comparator. The hypothesis advances only if it solves a consequential problem these alternatives leave unsolved, without creating dangers they avoid.
If the alternatives, taken as far as they can reasonably go, close the gap, the hypothesis fails. If a gap remains, the next question is whether a narrowly bounded constitutional system can fill it without becoming a dangerous concentration of power itself.
That second question is the problem of counter-agency. No body powerful enough to constrain frontier AI should control the evidence, the decision, the enforcement, the appeal, and the boundaries of its own jurisdiction. Every consequential exercise of authority must remain open to independent challenge.
What This Inquiry Is Trying to Produce
Stage One: Map the Powers
This is the current stage. SIRDA is identifying the powers that governing catastrophic AI risk requires, who holds each of them today, and where important powers are missing. The hypothesis and its comparators face the same hard problems and failure conditions. Stage One determines whether a genuine gap remains and whether government without territory deserves to advance.
Stage Two: Constitutional Design
If the hypothesis survives Stage One, SIRDA will convene an open design process that begins with the powers identified there and works out how each should be allocated, constrained, challenged, and withdrawn. Competing mechanisms stay visible throughout, so unresolved disagreement remains on the record.
Stage Three: Draft Constitution 0.1
If a viable architecture emerges, SIRDA will publish Draft Constitution 0.1 for public criticism. It would specify the powers the institution holds, the powers denied to it, how authority is divided, and how each exercise of authority can be challenged. Its test is demanding: it must decide consequential disputes that existing approaches cannot, while remaining more constrained than the danger it exists to address.
The Core Design Tension
The inquiry begins between two failures.
Too little common authority leaves catastrophic disagreements to bargaining among powerful actors, with rules that stay voluntary exactly when compliance matters most.
Too much common authority creates a concentration of power dangerous in its own right: an institution that expands its jurisdiction, suppresses dissent, falls to capture, or abuses emergency powers.
The constitutional task is to find out whether consequential collective action can exist between those failures.
The Chapters
Chapter 1: The Moment
Why advanced AI creates a governance problem with global reach, technical complexity, and potentially irreversible consequences.
Chapter 2: The Machine Problem
Alignment, control, evaluation, autonomy, and the technical problem of keeping increasingly capable systems within meaningful human oversight.
Chapter 3: The Human Coordination Problem
How people and institutions can coordinate around existential boundaries while continuing to disagree about values, politics, strategy, and the future.
Chapter 4: Counter-Agency
Why dangerous concentrations of agency require independent scrutiny, and why the institutions governing AI must themselves be governed.
Chapter 5: What Others Have Built
The existing technical, institutional, legal, and international work that this inquiry builds upon and challenges.
Chapter 6: The Missing Procedure
What happens after transparency and verification when legitimate actors still disagree about whether a consequential development should proceed.
Chapter 7: The Hypothesis: Government Without Territory
A capability-scoped model of distributed transnational governance and the boundaries that would define its mandate.
Chapter 8: The Hard Problems
The tiebreaker problem, collusion, disputed signals, jurisdictional thresholds, exit, outsiders, military autonomy, embodiment, cartel power, and other problems capable of defeating the proposal.
Chapter 9: Lessons From Government and Global Institutions
What emergency powers, capture, gamed tests, secret evidence, disabled courts, and international finance teach about building authority across borders, and where coordination without a sovereign stops.
Chapter 10: The Inquiry and Its Failure Conditions
How the project can be tested, challenged, amended, or abandoned when the evidence turns against it.
Continue the Inquiry
The chapters state the argument. The Open Questions isolate the constitutional problems that remain unresolved. The Failure Conditions define reasons the project should stop or change direction. The Ecosystem connects those questions to the people and institutions already doing relevant work.