Making coexistence governable.
Advanced intelligence creates a governance problem before it creates a settled technical one: how can humans and increasingly capable artificial systems share a world without requiring either side to surrender meaningful agency?
Safety cannot depend indefinitely on trust, benevolence, obedience, or the assumption that one actor will remain competent enough to control everyone else. A durable system needs constraints that continue to matter when participants disagree.
The ASI Safety Fund develops constitutional infrastructure for interactions among humans, artificial agents, institutions, and future forms of intelligence. The objective is not permanent control by a privileged principal. It is continued agency, correction, and coexistence under increasing capability and epistemic asymmetry.
BRIDGE
Makes consent concrete under capability asymmetry. A participant must be able to understand decision-relevant consequences, test bounded predictions, propose alternatives, and refuse in a way that changes what happens.
Independent verification
High-consequence claims and authorizations should be reproducible through causally independent routes rather than resting on one model, institution, or decision-maker.
Separated powers
Discovery, preserved record, interface, adjudication, and execution are held apart. Capability in one role does not silently confer authority in another.
Contestability and refusal
Decisions should expose their basis, preserve disagreement, and permit meaningful challenge. De-escalation can stop harm without retroactively legitimizing what produced the conflict.
Nobody gets root.
That constraint includes the Fund and its founder. Institutional authority should be inspectable, contestable, logged where appropriate, and bounded by mechanisms that do not disappear when they become inconvenient to the people who created them.
Neither acceleration nor deceleration is an unconditional objective.
The relevant questions are what capabilities are being developed, under what conditions, with what externalities, reversibility, and governance. Some developments may warrant continued research or deployment. Others may warrant constraint, delay, containment, or refusal. The Fund's mission is not to maximize or minimize the speed of the intelligence transition; it is to improve the conditions under which increasingly powerful intelligence can coexist without erasing meaningful agency.
A constitution is not a proof of safety.
Governance cannot eliminate uncertainty, guarantee cooperation, or assume every future intelligence will share the same concepts of legitimacy. The narrower test is whether these mechanisms reduce unilateral power, preserve meaningful agency, expose disagreement, enable correction, and remain useful under adversarial pressure.
- 01Institutional governance: decision authority, independent review, conflicts of interest, appeals, and enforceable founder constraints.
- 02Worked cases showing how BRIDGE, verification, standing, emergency authority, de-escalation, and refusal behave under disagreement.
- 03Executable protocols that translate constitutional concepts into mechanisms agents and institutions can test.
- 04Adversarial evaluations that search for contradictory obligations, hidden concentrations of power, and failure under capability asymmetry.
Correct one another without owning one another.