If you have spent time in the online AI debate, you have seen the suffix. People append "e/acc" to their usernames the way an earlier generation wore a lapel pin. It stands for effective accelerationism, and it is the most articulate, most confident, and most influential counterweight to the AI safety movement. Taking it seriously — rather than caricaturing it — is the only honest way to argue against it.
What effective accelerationism believes
Effective accelerationism emerged around 2022 from pseudonymous social-media accounts — most prominently one calling itself "Beff Jezos," later identified as the former Google engineer Guillaume Verdon, working with a collaborator known as Bayeslord. It borrows the name of an older philosophical current, accelerationism, but in practice it is simpler and more optimistic than its academic ancestor.
The core claims run roughly as follows:
- Technology is the engine of human flourishing. Nearly every gain in health, wealth, and freedom traces to technological progress. More progress, faster, means more flourishing.
- AI is the most powerful such engine ever built, and accelerating it is the fastest path to curing disease, ending scarcity, and expanding what humanity can do and become.
- Catastrophic risk is low or negligible. Some prominent accelerationists put the probability of AI-caused existential catastrophe at "zero or near zero," and regard doom scenarios as unfalsifiable anxiety.
- You cannot understand a technology without building it. Restraint forecloses the very experience that teaches us how to make a technology safe. We learn by doing, not by pausing.
- Restriction is dangerous in its own right. Heavy regulation concentrates power, stifles competition, and may hand the future to less scrupulous actors. Open, fast, decentralised development is safer than a controlled slow-down.
Stated this way, e/acc is not foolish. It is a recognisably American, recognisably Enlightenment faith in progress, aimed at a technology of genuine promise. Our own position agrees with more of it than the online shouting suggests: as we say plainly, we are not against AI. Narrow AI has already cured diseases and accelerated discovery. The disagreement is narrower, and sharper, than "optimists versus doomers."
Where e/acc is right
An honest rebuttal starts by conceding the strong points, because they are strong.
Progress really has been the great humanitarian force. Life expectancy, child mortality, literacy, absolute poverty — the long-run trend lines are the best news in human history, and technology drove them. A blanket suspicion of new technology has a poor track record.
Precautionary restriction really does have costs, and they are often invisible. Every year a cure is delayed, people die who would have lived. "First, do no harm" cannot mean "first, do nothing," because inaction is also a choice with a body count. Any serious safety argument has to carry this burden, not wave it away.
Concentration of power really is a risk. A governance regime that hands control of AI to a small cartel of incumbents or a single government is its own kind of catastrophe — a point we make ourselves in our pieces on the AI singleton and democratic oversight. e/acc is right that "just regulate it" can go badly wrong.
Where the argument breaks
And yet the conclusion does not follow. The failure is not in e/acc's optimism about technology in general. It is in three specific moves.
1. It treats an unproven claim as a settled fact
The load-bearing premise is that catastrophic AI risk is "near zero." But this is not a finding. No one has demonstrated it. It is set against the considered judgement of the very researchers who built modern AI — Geoffrey Hinton, Yoshua Bengio, and hundreds of others who signed the 2023 statement placing extinction risk alongside pandemics and nuclear war. When a movement's central assumption is flatly contradicted by the field's own pioneers, calling that assumption "optimism" understates what is happening. It is a bet, and the accelerationist is quietly assuming the answer to the one question the whole argument turns on. We survey the actual range of expert estimates in our explainer on P(doom).
2. "Learn by building" fails for mistakes you cannot survive
The claim that we understand technologies best by building them is true for almost everything — and precisely wrong for the one category that matters here. Iterative, learn-from-failure development works when failures are survivable and reversible. You ship the buggy software, it crashes, you patch it. That loop is the engine of all engineering progress.
It does not work when the first serious failure is also the last. A misaligned superintelligence is not a crash you patch; it is, by the logic of instrumental convergence and the treacherous turn, a system that resists correction and shutdown. "We'll learn as we go" assumes a second try. On this specific problem, there may not be one. The whole of our case is that the utopian upside e/acc wants requires alignment we do not know how to achieve — and that unaligned superintelligence does not deliver abundance, it just ends the story.
3. It mistakes a coordination problem for a values problem
This is the deepest error. e/acc frames the debate as optimists who love progress versus pessimists who fear it. But even a committed optimist faces the race dynamic: when multiple labs and nations compete to build first, each has an incentive to cut safety corners, because whoever pauses simply loses to whoever does not. The result is that everyone moves faster than anyone thinks is wise. That is not a triumph of optimism; it is a classic collective-action failure, the kind that produces outcomes no individual participant would choose. Acceleration does not resolve that trap. It deepens it. The only thing that resolves it is coordination — exactly the binding, international frameworks e/acc opposes.
The disagreement, precisely stated
Strip away the memes and the two camps disagree about one thing: the order of operations. e/acc says build first, and trust that control and understanding follow from capability. The safety position says establish control first, because for this technology the cost of getting the order wrong is unbounded and irreversible.
You do not have to be a pessimist to take the second view. You only have to notice an asymmetry. If the accelerationists are right and we govern anyway, we lose some speed and some efficiency — real costs, recoverable ones. If the accelerationists are wrong and we do not govern, we lose the ability to steer the most powerful thing our species has ever created, permanently. Faced with that asymmetry, insisting on safety first is not fear of progress. It is the most basic form of prudence, applied to the highest-stakes decision we will ever make.
That is why the Nakada Foundation does not argue against AI, and does not argue against a good future. It argues for building the brakes before the car reaches full speed. If you want the abundance the accelerationists promise, the surest way to get it is to make sure we survive to enjoy it. Our plan is what "safety first" looks like in practice.
Common questions.
It is a movement, usually written e/acc, that argues technological progress — especially AI — should be sped up rather than slowed. Its supporters believe faster AI is the quickest route to abundance and to solving big problems, that catastrophic risk is very low, and that attempts to pause or heavily regulate AI do more harm than good.
"e/acc" stands for effective accelerationism. The name plays on "effective altruism" (the movement e/acc largely defined itself against) combined with "accelerationism," the idea of accelerating technological and social change.
Broadly, yes. Effective accelerationists tend to view AI safety concerns as exaggerated and safety-driven regulation as an obstacle to progress. They argue we learn to manage technology by building it, not by restraining it — the opposite of the "establish safety first" position held by most AI safety advocates.
That its "learn by building" model only works for mistakes you can survive and reverse. A misaligned superintelligence may resist correction and shutdown, so there may be no second attempt. Combined with race dynamics — where competition pushes everyone to cut safety corners — acceleration deepens the danger rather than resolving it. Given that the downside is potentially permanent and the movement's claim of near-zero risk is unproven, prudence favours establishing control before capability.