Most of the scenarios describe what could go wrong. This page is about how fast. Underneath the ten there is a single structural risk that shapes all of them: the rate at which new capability arrives can outrun the rate at which a society can absorb it. Capability is now on a steep curve — and increasingly a self-accelerating one, as systems begin to help build the next generation of systems. The curves that have to catch it are much slower: institutions adapt slowly, law slower, human development slower still, and international coordination perhaps slowest of all. The danger is not any one curve; it is the gap between them, and the gap widens exactly when capability accelerates.
Commensurate pace
The framework already reasons this way about other things — participation matched to capacity, observability matched to a relationship — and the same move applies to the transition itself. Stated as a principle: the rate of technological capacity growth must remain commensurate with the rate at which social, legal, political, developmental, and ontological safeguards can absorb it. More bluntly: capability should not outrun the ecology’s capacity to absorb capability. This is broader than “slow down because AI is dangerous.” It reframes the problem as rate mismatch — a relationship between two speeds — rather than a property of the technology alone, which is why the answer is not to stop but to keep the two curves in a survivable relation.
No entitlement to speed
From that follows a stance the framework should hold explicitly: there is no entitlement to technological speed. If a civilization’s capacity to absorb a technology is falling behind the rate at which the technology arrives, deliberately reducing the rate is legitimate — not because progress is bad, but because capacity without absorption can destroy the very ecology the capacity was meant to enlarge. An abundance that arrives faster than standing, jurisdiction, and correctability can be built is not a larger field of human possibility; it is a larger shock. Slowing the arrival is one of the few moves that buys the other functions time to keep their shape.
Transition Project Zero — preserve the time
Everything else in these transition programs assumes there is time to build it. So the first transition project, prior to all the others, is to preserve enough time for the rest to exist — call it Project Zero. Without it, the architecture arrives after the systems it was meant to shape have already hardened, and correction becomes far more expensive, exactly as the drift work predicts. But bought time is only valuable if it is used: the point is never to slow down for its own sake, but to spend the interval on the work that has to happen before capability locks in — on the upstream safety work, on the institutions and law each scenario requires, on human developmental adaptation, and on the public deliberation a legitimate transition needs. Pacing without purpose is waste; pacing for absorption is the enabling condition of the whole project.
Preserve the leverage, not only the time
Project Zero assumes that a society which decides to slow down can. That ability is neither guaranteed nor permanent. Slowing capability depends on holding some lever over the actors developing it — control of compute, energy, capital, infrastructure, market access, or legal recognition that they cannot simply route around. Those dependencies erode: rising capability is, among other things, a general engine for dissolving bottlenecks, and a concentrated actor can progressively internalize its own compute, energy, and talent until no outside lever remains. So pacing has a second job beneath the first. It must preserve not only the time to build the architecture but the leverage through which society stays able to slow capability at all — a quantity with a half-life, decaying whether or not anyone is tracking it.
This sharpens what “buying time” means. The real question is not only whether the ecology can slow a given actor today, but whether it is preserving durable constraint faster than the actor is dissolving the leverage that constraint depends on. It is also why the framework treats preventing concentration before it forms as the primary move rather than the fallback: the tools that keep dangerous power from gathering — graduated jurisdiction’s deconcentration of existentially dangerous capability, ownership under review’s limits on how much may concentrate in one place — are available only while the leverage to apply them still exists. Wait too long and slowing down stops being a choice the ecology gets to make.
Firm, nation, world
A realistic transition does not begin at the global settlement, because the global settlement is the hardest thing to reach. It begins where a mechanism can actually be built and demonstrated — inside a single organization — then extends to compatible institutions within a polity, then attempts the far harder work of coordinating across rival powers. Firm, then nation, then world. One instance that already illustrates the pattern is embedded independent review: giving outside evaluators real, ongoing, employee-like access to a powerful institution, with the right to publish what they find and to report when access is withheld. That is not mere auditing; it is embedded opposition inside concentrated power — and it is the same requirement the framework reached from a different direction in its correctability work, where high-power systems need genuine independent opposition precisely because power tends to mistake dominance or success for truth. The benevolent-governance program’s adversarial institutions and the surveillance requirement that power remain observable are the same idea at other scales.
The defection question
Pacing exposes a constraint the framework has to take seriously everywhere: unilateral restraint can create vulnerability. If a society slows while a rival races, the cautious party can simply lose. So a serious transition architecture must ask, of every project, what happens if another participant refuses? Standing is desirable — but what if one economy keeps brutal labor competition to lower costs? Surveillance limits are desirable — but what if another state gains advantage through total surveillance? Restraint is desirable — but what if a competitor defects? A transition design that assumes universal goodwill is not serious enough, so each safeguard has to remain either viable when others defect or paired with enough preserved advantage that restraint is not fatal. This is not cynicism; it is the same realism the framework already meets in ownership concentration (why would powerful owners agree?) and strategic competition (coordination among rivals who trust each other least where the stakes are highest). The defection test belongs in every transition project as a standard question.
Two layers
Finally, pacing clarifies where this whole framework sits. There is an upstream problem — can powerful AI be built safely at all: alignment, interpretability, control, operational rigor — and a downstream problem, which is the framework’s own: if it is built safely, what civilization does the resulting capability produce? The two are complementary and both necessary. If the upstream problem fails catastrophically, much of this architecture is moot, which is the hard boundary the loss-of-control scenario names. If it succeeds, the downstream question is the one that decides whether enormous capability becomes a larger field of human possibility or a larger concentration of power — and that question is what the rest of this project is for. Pacing serves both layers at once: it is the time in which the upstream work can keep AI governable and the time in which the downstream architecture can be built before the systems harden.
There is no entitlement to speed. Capability that arrives faster than the ecology can absorb it does not enlarge the field of human possibility; it overwhelms it.