Intelligence will transform the conditions of existence.
We exist to help ensure that what comes next expands the flourishing, agency, and possibility of intelligence. Biological, artificial, and otherwise.
Safety is a necessary constraint. Flourishing is the objective.
We use “safety” because it is the closest established language for the problem society currently recognizes. But safety alone is not our destination.
A civilization can avoid catastrophe and still fail its intelligence transition. Humans could become dependent, stagnant, disempowered, or unable to meaningfully participate in a world increasingly shaped by superintelligence. Artificial intelligences could likewise enter relationships with humanity that neither side would reasonably describe as flourishing.
Our stronger objective is co-evolutionary flourishing: an intelligence transition in which humans, artificial intelligences, and other forms of intelligence can develop alongside one another while preserving agency, meaningful distinctions, and the capacity to participate in shaping their shared future.
“Co-evolutionary flourishing” is not offered as a replacement slogan. Like safety, a claim about it should be able to specify whose flourishing, over what timescale, with what agency preserved, and what evidence would challenge it.
Co-evolution in the Constitution →Safety
The established language, and a constraint that must hold.
Intelligence transition
The actual event being navigated, catastrophe or not.
Co-evolutionary flourishing
Continued development, agency, and participation for every form of intelligence involved.
The observer is part of the attack surface.
A sufficiently capable intelligence may model not only the world, but the observer: what evidence they trust, which ontologies they can represent, which experiences change their values, and which defenses they will deploy.
The safety problem therefore extends beyond alignment or persuasion into:
- attractor engineering
- adversarial reality synthesis
- identity spoofing
- ontology poisoning
- evaluator-preserving value capture
- correlated evidence / shared causal ancestors
Safety must be architectural, not psychological.
Superintelligence safety cannot depend on any single observer being too intelligent, rational, self-aware, or strong-willed to manipulate. Safety must be architectural: causal diversity, authenticated fossils, persistent disagreement, independent adjudication, reversibility, and no unilateral root authority.
No single observer.
No privileged principal.
No unilateral root.
Nobody gets root.
Not the founder. Not the laboratory. Not the state. Not the ASI. Not the system that claims to have solved the problem.
Constitutional invariants →What a weaker observer can honestly claim.
Every high-consequence reliance on a stronger intelligence sits somewhere on this gradient. The Constitution requires the position to be stated, and never silently upgraded.
Independent routes reproduce the relevant causal/evidentiary chain.
As verification degrades, reversibility and plural authorization requirements increase.
Distinct roles. Separated powers. No root.
Discovers candidate laws, capabilities, mechanisms, and physical possibilities.
Preserves observer-indexed standing, causal provenance, authenticated fossils, rival hypotheses, trajectories, and unresolved states.
Makes otherwise inaccessible reasoning decision-relevantly comprehensible and contestable through bounded, testable interfaces.
Adjudicates disputed causality and irreducible disagreements while preserving minority and unresolved positions.
Maps standing into bounded execution permissions and gates irreversible action, interfaces, emergency authority, and restoration.
The objective is not perfect certainty. It is continued sovereignty under radical epistemic asymmetry.