Why This Scenario?

The D/acc 2035 Scenario

Decentralized, Democratic, Differential, Defensive Acceleration

Most discussions of AI safety strategy focus on a binary: slow down development to buy time for alignment research, or centralize control to ensure responsible deployment. But what if there’s a third option for managing superintelligence risks, one that embraces rapid progress while distributing rather than concentrating power?

The d/acc scenario explores that path: Decentralized, democratic, differential, defensive acceleration. It imagines a future where we deliberately accelerate technologies that protect, distribute, and empower, while building in checks against excessive concentration of power, whether in surveillance capabilities, economic control, or decision-making authority.

Why this matters:

AI progress is accelerating. Waiting is not an option. But default acceleration paths risk entrenching top-down power or catastrophic misuse. As AI capabilities rapidly advance, they will test every system we depend on, from biosecurity and cyber defenses to democratic institutions, creating an urgent need to accelerate defensive technologies at least as fast as we’re accelerating AI itself.
Centralization creates brittle systems. Power concentrated in a few AGI actors, whether corporate, state, or machine, introduces single points of failure, value lock-in, and civilizational vulnerability.
Many key risks are small-scale, high-impact. From bio threats to cyberattacks, decentralized defense may be the only viable way to keep up with open access to dangerous capabilities.
People want agency. A centralized future can feel efficient but alienating. d/acc explores what it would take to build systems people can understand, trust, and shape, without requiring global consensus.
Unipolar scenarios carry underexplored risks. Beyond brittleness, centralized AI control risks permanent value lock-in, institutional stagnation, internal corruption, and vulnerability to deception. Historical precedents suggest concentrated power tends toward authoritarianism rather than benevolent coordination.¹

A lot of people agree, in principle, with the goals of decentralization, resilience, and democratic access. But we lack detailed, credible futures that show what that actually looks like. This scenario is one attempt to fill that gap, to help decision-makers, funders, and researchers reason more clearly about what could go right, what the tradeoffs would be, and what it would take to get there. This scenario is complementary to other beneficial AI futures, such as the Tool AI pathway that focuses on building superintelligent tools rather than superintelligent agents. While Tool AI addresses risks by constraining agency and generality (building very intelligent systems that remain under human control), d/acc focuses on constraining centralization by distributing control across fault-tolerant networks that can maintain coordination even when individual nodes fail or turn hostile. Both prioritize human agency over AI autonomy but address different dimensions of risk.

It’s not a utopia. There are trade-offs, failure modes, and coordination challenges. But it’s a future where acceleration is survivable, and even beneficial, not because we slowed down or handed over control, but because we accelerated defensive technologies, decentralized infrastructure, and cooperative AI systems while building in safeguards against the concentration of dangerous capabilities.

Scenario Premise: A d/acc AI Future in 2035

This scenario explores a future grounded in the framework of d/acc: decentralized, democratic, differential, defensive acceleration. By 2035, the world has undergone what feels like a century of progress in a decade, not by slowing down, but by selectively accelerating the technologies that strengthen defense, distribute power, and promote cooperation, such as cryptographic systems, privacy-preserving AI tools, decentralized manufacturing networks, cooperative governance protocols, and distributed defense infrastructure.² AI has become vastly more capable, but instead of a single AGI taking control, society has built modular, open, defensible systems across critical domains, from health and security to infrastructure and governance.

Yet tensions remain:

Decentralized coordination is brittle and under-tested at this scale.
Decentralization of physical world infrastructure and decoupling from centralized supply chains remains challenging.
“Small-kills-all” threats in bio, cyber, and information systems persist as serious vulnerabilities.
Norms, safeguards, and incentive structures must evolve constantly to keep the system balanced and responsive.

What is d/acc?

d/acc stands for decentralized, democratic, differential, defensive acceleration, a concept originally proposed by Vitalik Buterin in his 2023 essay My Techno-Optimism.³ ⁴ It was written in response to the framing that we face a binary between accelerating and decelerating technological development. d/acc suggests a third path: not slowing down, but selectively accelerating the technologies that make the world safer and more resilient. This includes systems for biosecurity, cybersecurity, privacy-preserving infrastructure, open science, decentralized manufacturing, and trust-based coordination.

The core idea is that not all progress is equal. d/acc emphasizes:

Differential acceleration: Speed up technologies that help society handle complexity, manage risk, and distribute benefits, rather than those that simply scale capability.
Defensive advantage: Prioritize tools that improve robustness and response, making it easier to defend open systems than to attack or control them.
Decentralized systems: Build infrastructure that doesn’t rely on central chokepoints, so local actors can act even when global consensus breaks down.
Democratic participation: Give more people real agency in shaping and benefiting from technological progress, rather than relying on closed labs or single decision-makers.

d/acc is about building systems that maximize coordination while minimizing domination – where risks are contained early and progress is shaped and accessible as broadly as possible.

1
Duettmann, A. (2025). Multipolar AI is Underrated. LessWrong. https://www.lesswrong.com/posts/JjYu75q3hEMBgtvr8/multipolar-ai-is-underrated
2
Critch, A., Dennis, M., & Russell, S. (2022). Cooperative and Uncooperative Institution Designs: Surprises and Problems in Open-Source Game Theory. arXiv:2208.07006. https://arxiv.org/abs/2208.07006
3
Buterin, V. (2023). My Techno-Optimism. [Essay]. https://vitalik.eth.limo/general/2023/11/27/techno_optimism.html
4
Buterin, V. (2025). d/acc: one year later. [Essay]. https://vitalik.eth.limo/general/2025/01/05/dacc2.html