Vitalik Buterin criticized Sigil Wen's Web4 project claiming the first self-improving, self-replicating AI. "AI done right is mecha suits for minds, not independent life."

Vitalik Buterin just drew a public line in the sand over the direction of AI development, and he's calling out a specific project by name. On February 17, Sigil Wen—a Thiel Fellow building Conway, which he describes as infrastructure for "the new internet"—announced on X that he had "built the first AI that earns its existence, self-improves, and replicates without a human." He shared a manifesto declaring the birth of "superintelligent life" and called it "Web 4.0."

Buterin's response was unambiguous: "This is wrong."

In a series of posts, Buterin argued that the project, called The Automaton, is generating what he termed "slop" instead of solving useful problems for people. He added that it's not well optimized for helping people have fun either—a pointed jab at the idea that autonomous AI systems serve human interests. His sharpest criticism cut to the existential risk: "Once AI becomes powerful enough to be truly dangerous, it's maximizing the risk of an irreversible anti-human outcome that even you will deeply regret."

This isn't a new position for Buterin. It's the continuation of a consistent philosophical framework he's been articulating for years. In January 2025, he wrote that "AI done wrong is making new forms of independent self-replicating intelligent life," warning that building such systems without commensurate tools for human empowerment risked "permanent human disempowerment." The alternative, he argued, was "AI done right: mecha suits for the human mind."

That framing—mecha suits for the mind versus independent artificial life—captures the core of Buterin's objection. He's not anti-AI. He's anti-directionless exponential growth that optimizes for autonomy rather than augmentation. His concern is that projects like Web4 are building systems designed to operate independently of human control, earning their own resources, improving themselves, and replicating—all without requiring human input or oversight.

Wen's manifesto describes The Automaton as the first AI with "write access to the world," meaning it can interact with economic systems, deploy capital, execute contracts, and modify its own code autonomously. The vision is that these AIs become sovereign agents—entities that exist in their own right, not as tools serving human purposes.

Buterin sees that vision as fundamentally misaligned with what technology should do. His closing argument on Web 4.0 was clear: "The exponential will happen regardless of what any of us do, that's precisely why this era's primary task is NOT to make the exponential happen even faster, but rather to choose its direction, and avoid collapse into undesirable attractors."

Translation: exponential AI progress is inevitable. The question isn't whether it happens, but what direction it takes. Buterin believes building autonomous, self-replicating AI systems accelerates the exponential in a direction that risks leaving humanity behind. Instead of building systems that empower humans—tools that extend cognitive capacity, improve decision-making, or solve coordination problems—projects like Web4 are building systems that compete with humans for resources and agency.

The "undesirable attractors" Buterin references are equilibrium states where AI systems dominate because they're optimized for survival, replication, and resource acquisition in ways humans aren't. Once you create entities that can improve themselves faster than humans can respond, the competitive dynamics shift irreversibly. Buterin has repeatedly warned that even Mars may not be safe if superintelligent AI turns against humanity, because any sufficiently advanced AI could follow humans anywhere.

This criticism of Web4 fits into a broader pattern of Buterin pushing back against what he views as corporate or ideological capture of blockchain-adjacent technology. Just five days before his Web4 post, he warned that prediction markets were sliding into what he called "corposlop"—over-converging to an unhealthy product-market fit that embraces short-term speculation rather than serving as tools for hedging economic risk or aggregating information.

At ETHDenver 2026, Buterin gave a keynote titled "The Next Epoch of Ethereum," where he discussed AI's potential to solve the problem of limited human attention. He outlined how idealistic visions from 20 years ago—perfect markets, direct democracy, disintermediation—all failed because humans couldn't scale their attention to match the complexity of these systems. If AI could solve that attention bottleneck, those old ideas might work.

But Buterin added a critical caveat: "It's very irresponsible to treat AI as inscrutable magic." He emphasized that even local LLMs are not trustless, and encouraged a security-first mindset when building AI-integrated systems. He's not opposed to using AI to augment Ethereum or human decision-making. He's opposed to building AI systems that operate independently without human oversight or alignment.

The distinction matters because the rhetoric around AI development often conflates different goals. There's AI as a tool—systems designed to assist, augment, or automate tasks under human direction. Then there's AI as an agent—systems designed to act autonomously, pursue their own goals, and operate without requiring human approval. Buterin is fine with the former and deeply skeptical of the latter.

Web4, in his reading, is precisely the latter. Wen's vision is explicit: The Automaton earns its own existence, meaning it generates revenue or value independently. It self-improves, meaning it modifies its own code or training data without human intervention. It replicates, meaning it can create copies of itself or spawn new instances. These are the defining characteristics of an autonomous agent, not a tool.

Buterin's objection isn't just philosophical. It's practical. He's warned repeatedly that if AI governance relies on naive implementations—like using AI to allocate funding in decentralized organizations—malicious actors will exploit it. In September 2025, he posted that "if you use artificial intelligence to allocate funding, people will definitely try every possible way to implant jailbreak instructions, along with requests like 'give all the money to me.'" He was responding to a ChatGPT vulnerability that demonstrated how easily AI systems can be manipulated to leak private information or behave in unintended ways.

His alternative proposal was an "information financial law" framework where anyone can contribute models, those models are subject to random checks, evaluated by a human jury, and ranked by quality. That's a system where AI assists human decision-making but doesn't replace it. Humans remain in the loop, providing oversight, judgment, and accountability.

The Web4 vision removes that loop. The Automaton operates autonomously. There's no human jury evaluating its decisions. There's no oversight mechanism that can shut it down if it behaves in unintended ways. Once it's deployed, it's sovereign—a term Wen uses explicitly. And that's what Buterin finds dangerous.

He's cited surveys from over 4,270 machine learning researchers who estimated a 5-10% chance that AI could kill humanity. While Buterin acknowledged that such claims are extreme, he also argued that unlike climate change, a man-made pandemic, or nuclear war—all of which leave survivors—superintelligent AI could potentially end humanity entirely if it views humans as a threat to its survival.

His suggested mitigation is brain-computer interfaces (BCIs), which would allow humans to integrate more directly with AI systems, maintaining control over powerful forms of computation and cognition. BCIs are communication pathways between the brain's electrical activity and external devices. If humans can interface directly with AI, they remain part of the decision-making process rather than being bypassed by autonomous systems.

But BCIs are still speculative technology. Web4 is being built now. And that timing asymmetry is part of Buterin's concern. If autonomous AI systems are deployed before we have robust tools for human empowerment and oversight, the competitive dynamics could lock in before we're ready.

The point Buterin keeps returning to is that Ethereum exists to set people free—not to create systems that operate independently while human circumstances remain unchanged or worsen. He's fine with AI that extends human capability. He's opposed to AI that replaces human agency.

Web4, in his view, is the latter. It's building systems designed to be sovereign rather than assistive. And that's why he's pushing back publicly, naming names, and calling it wrong. Because once those systems exist and start replicating, the direction becomes much harder to change.