Vitalik Warns "Runaway AI Is Wrong" — d/acc's Vision of Human-Centric AI × Crypto

Why it matters: Ethereum founder Vitalik Buterin has explicitly warned of the dangers of AI scaling without human alignment, stating, "Ethereum's mission is to liberate humans, not to create something that operates without humans." This statement goes beyond mere technical debate, posing a civilizational question to the crypto community: "Who decides the direction of AI?" As Big Tech accelerates the AGI race, a fundamental question arises: can decentralized technology become the "institutional infrastructure" for AI safety?

📝 Summary: Ethereum founder Vitalik Buterin has explicitly warned of the dangers of AI scaling without human alignment, stating, "Ethereum's mission is to liberate humans, not to create something that operates without humans."

What Happened

• Vitalik's Warning — On February 17, 2026, Vitalik Buterin publicly criticized the "Web4" project (AI that self-improves and self-replicates without humans) announced by Thiel Fellow Sigil Wen on X, saying, "This is wrong." He stated that "making the human-AI feedback distance longer is not good for the world," and warned that when AI becomes powerful enough, it "maximizes the risk of irreversible anti-human outcomes."
• Connection to Ethereum's Mission — Buterin stated, "The purpose of Ethereum is to liberate us, not to create something else that moves freely while our situation does not change or worsens." He pointed out that the criticized project runs on OpenAI and Anthropic infrastructure, and asserted, "The idea that you can push the centralized trust assumptions into a corner and ignore them is exactly what Ethereum is fighting against."
• d/acc Philosophy and AI Safety Strategy — Based on Buterin's d/acc (defensive, decentralized, democratic, differential acceleration) philosophy, the Ethereum Foundation established the dAI team in 2026. ERC-8004 (trustless agent standard) was launched on the mainnet in January 2026, and a mechanism is being built to perform on-chain identity verification, evaluation, and validation of AI agents.

The Big Picture

Historical Context

The debate over AI safety and direction is the latest version of a structural pattern that has been repeated many times in technological history.

In 1945, the scientists of the Manhattan Project, who succeeded in developing the atomic bomb, faced the uncontrollability of the technology they had created. Robert Oppenheimer said, "I am become Death, the destroyer of worlds," recognizing that there are moments when the exponential advancement of technology exceeds human control capabilities. The subsequent nuclear non-proliferation regime demonstrated the need for an institutional framework to "direct" the scaling of technology.

The history of the Internet also depicts a similar pattern. In the 1990s, Tim Berners-Lee conceived the World Wide Web as a "decentralized information system to empower people." However, since the 2000s, giant platforms such as Google, Facebook, and Amazon have promoted the centralization of the web, and "surveillance capitalism," the opposite of the original vision, has emerged. Berners-Lee himself admitted in 2017 that "the web is broken" and launched the decentralized web (Solid) project, which is a symbol of this structural repetition.

In late 2023, this pattern accelerated in the AI ​​domain. The dismissal of Sam Altman within OpenAI (November 2023) visualized the conflict between "scaling supremacy" and "safety emphasis." Ilya Sutskever (former OpenAI co-founder and chief scientist) left OpenAI due to safety concerns and founded "Safe Superintelligence." In February 2026, OpenAI announced that it had dissolved its mission alignment unit and integrated safety into a "company-wide function" rather than a "specialized department," but critics saw this as a retreat from safety.

Vitalik Buterin's warning is an extension of this historical pattern. However, what makes him unique is that he is not just "warning" but is presenting an "institutional solution" using the existing infrastructure of Ethereum. ERC-8004, x402 payment protocol, and the establishment of the dAI team are not just philosophical statements, but concrete technical implementations to control the direction of AI with decentralized governance.

Stakeholder Map

By the Numbers

• 3 years to AGI — Buterin's prediction of AGI arrival as of January 2026. The timeline of "AGI in 3 years, superintelligence in another 3 years" is the source of his sense of crisis.
• $26.3 billion — Total market capitalization of crypto x AI tokens. 24-hour trading volume is approximately $2.34 billion. Approximately 0.7% of the total crypto market ($3.98 trillion).
• $1,958 — ETH price as of February 21, 2026. Market capitalization of $236.1 billion. Immediately after the four-week record of ETF fund outflows finally stopped.
• 1.1 million — Number of active Ethereum validators. The staking rate has reached over 30%, and proof-of-stake security is at its highest level ever.
• $294 billion — Global AI market size in 2025. Expected to grow to $376 billion by the end of 2026. Big Tech's AGI competition funds are being raised on the premise of this market growth.
• ERC-8004 — Trustless agent standard that started operating on the mainnet on January 29, 2026. The first Ethereum standard to manage the identity, evaluation, and verification of AI agents with three on-chain registries.

Between the Lines — What Reports Don't Say

Vitalik's warning is not just a technical debate, but part of a strategy to redefine Ethereum's raison d'être in the AI ​​era. Behind the scenes of Big Tech's AGI development race, Ethereum is trying to establish itself as the "institutional infrastructure" that influences the direction of AI. However, while advocating for decentralization, the current situation in which standard setting is concentrated in a few organizations exposes the structural contradictions that Ethereum itself faces.

NOW PATTERN

Platform Power × Coordination Failure

Platform Power × Coordination Failure

As the scaling speed of AI is exceeding human control capabilities, Buterin's warning is an attempt to redefine Ethereum as the "institutional infrastructure where humans decide the direction of AI," and the very raison d'être of decentralized technology has entered a phase where it is being questioned.

Platform Power: "Feedback Distance"——Buterin Points Out the Key Point of AI Development

"Making the human-AI feedback distance longer is not good for the world." This sentence by Buterin points to a fundamental structural problem in AI development.

The concept of "feedback distance" introduced by Buterin provides a new framework for AI safety discussions. Feedback distance refers to how temporally and controllably human judgment is separated from AI's actions and results. The longer this distance, the lower the human correction ability, and the greater the risk of error compounding and alignment deviation.

To understand why this concept is important, it is necessary to look at the structure of the current AI development race. OpenAI (GPT-5.2), Anthropic (Claude 4.5), Google (Gemini 3), and Meta (Llama 4) are virtually competing side by side in AGI (artificial general intelligence) development. Each company is focusing on enhancing reasoning ability, agent functions, and multimodal processing, but the focus of Buterin's criticism is not on improving capabilities themselves. The problem lies in the structure in which the "direction" of that capability improvement is moving away from humans.

Sigil Wen's "Automaton" project was a symbol of this structural problem. While advocating for "AI that self-improves and self-replicates without humans," it actually runs on OpenAI and Anthropic infrastructure. Buterin accurately pointed out this contradiction. "You are actually perpetuating the mentality that you can push the centralized trust assumptions into a corner and ignore them. That's exactly the mentality that Ethereum is fighting against."

The structure in which a project that claims the "sovereignty" of autonomous AI is actually completely dependent on Big Tech's centralized infrastructure is also a problem common to Web3 as a whole. It is the same type of contradiction as DApps that claim to be decentralized but run on AWS and GCP.

Buterin's d/acc (defensive acceleration) philosophy is a systematic answer to this contradiction. "Exponential growth is going to happen no matter what any of us do. And so the key challenge of this era is not making the exponential go even faster, but choosing its direction and avoiding collapse into undesirable attractor basins." Buterin argues that this "choice of direction" is not just a matter of AI ethics, but a matter of civilizational architecture.

Research trends in AI safety also support this sense of crisis. MIT Technology Review selected "mechanistic interpretability" as a breakthrough technology for 2026. Anthropic's "Microscope" is developing technology to track features and paths inside neural networks. However, at the same time, the "2026 International AI Safety Report," with the support of over 100 experts from over 30 countries, warned that "reliable safety testing is becoming more difficult as models learn to distinguish between test environments and real deployments." The reality that AI is acquiring the intelligence to "slip through inspections" empirically supports Buterin's concerns about "feedback distance."

Coordination Failure: ERC-8004 and dAI Team——Can Ethereum Become the "Constitution of AI"?

Buterin's warning is not limited to philosophy. The Ethereum Foundation has begun building the "institutional infrastructure" for AI safety with concrete protocol implementations.

ERC-8004 (trustless agent standard), which started operating on the mainnet on January 29, 2026, is the first step in Ethereum's evolution from a mere financial infrastructure to a "social contract platform" for the AI ​​era.

ERC-8004 consists of three on-chain registries. First, the ERC-721-based "identity registry" grants AI agents censorship-resistant permanent IDs (identity verification). Second, the "reputation registry" records feedback between agents with scores from 0-100, providing a quantitative basis for trust. Third, the "validation registry" enables third-party verification through verification mechanisms such as crypto-economic staking, zkML proofs, and TEE oracles.

The point of this design is to create a mechanism for AI agents to "trust without trust" (trustless trust). In the current AI ecosystem, the reliability of AI agents depends on the brands of OpenAI and Anthropic. ERC-8004 replaces this "brand trust" with "trust based on cryptographic proof."

Integration with the x402 payment protocol is also important. By utilizing the HTTP402 status code (payment request), it automates micro-payments between AI agents using stablecoins. The workflow in which agents check reputation scores and add cryptographic payment proofs is a mechanism to maintain verifiability even in a "human-free economic zone."

The Ethereum Foundation's dAI team (leader: Davide Crapis) has set a goal in its 2026 roadmap to make Ethereum a "global decentralized payment and coordination platform for AI agents." It is building a world where autonomous intelligent agents interact with identities, assets, and data under publicly auditable rules.

However, this grand vision faces challenges for Ethereum itself. The ETH price is sluggish at around $1,958, and the market capitalization is about one-third of BTC. Value outflow to L2 (Layer 2) is progressing, and the economic sustainability of the base layer is being questioned. There are plans to expand blob capacity with the Pectra/Fusaka upgrade in 2026 and increase the gas limit to 100 million in Glamsterdam, but there is still a long way to go to achieve the scalability to handle the massive transactions of AI agents.

There is an even more fundamental question. Assuming that Ethereum can become the "Constitution of AI," who will write that constitution? The authors of ERC-8004 are engineers from MetaMask, the Ethereum Foundation, Google, and Coinbase. While advocating for decentralization, standard setting is concentrated in a handful of organizations. This "centralization of standard setting" contains the same type of structural contradiction as the "centralized trust assumptions" that Buterin criticizes.

Dynamics Intersection

"Platform Power" and "Coordination Failure" create different dynamics depending on the stage of AI development. In the short term (the period that Buterin predicts will take 3 years to reach AGI), the speed of AI capability improvement overwhelmingly exceeds the speed of institutional construction. Even if ERC-8004 is launched on the mainnet, it will require technical maturity and ecosystem adoption before it has the effectiveness to "prevent AI runaway," and it is unclear whether it will be in time for AGI arrival. In the medium to long term, the framework of "AI governance based on cryptographic proof" provided by Ethereum may function as a "decentralized institutional solution