Crypto: Vitalik Buterin wants to use artificial intelligence to improve DAOs

DAOs dream of a world where decisions are made “by the crowd”. In practice, the crowd yawns. And Vitalik Buterin has just pointed out the real crux: it's not (only) a problem with periods, it's a problem with attention. Too many votes, too many topics, too much technicality in crypto… and in the end, a minority decides while the majority lets it pass.

The real bug of DAOs: not democracy, exhaustion

Buterin speaks of “limits of human attention”. It's almost banal, and yet it's brutally precise: the average crypto holder doesn't have the time to read 12 proposals, compare three audits, understand the consequences on cash flow and anticipate second-order effects. Result: abstention becomes a hidden functionality of the system.

It is often estimated that the average participation of DAOs is around 15% to 25%. At this level, you no longer have community crypto governance: you have a club, with an open door and few people passing through. This vacuum creates two toxic effects: de facto centralization (always the same people who vote) and the slowness of decisions (or worse: decisions taken “by default”).

And when vigilance drops, the scenarios become darker. Buterin recalls the classic blind spot: the attack on governance. A bad actor can amass enough voting power, propose something destructive, and push it through while the crypto community… isn’t looking. This isn't science fiction, it's the kind of flaw that's just waiting for good timing.

Delegating is practical… but it cuts the thread

The historical “solution” is delegation: you entrust your voice to a representative, then you move on. Vitalik speaks of a “disempowering” system. In short, you press a button and you lose all fine influence. Your delegate may be competent, but you no longer have the steering wheel, only the horn.

The problem is not that delegates exist. This is because they become a bottleneck. A few profiles end up concentrating attention, relationships, information… therefore power. And even without corruption, governance begins to resemble professional politics: a few speak, the rest “applaud” or “ignore”.

This is where Buterin offers an alternative stranger, almost intimate: instead of delegating to a person, delegate to an AI assistant… but an assistant who looks like you. Not a “party advisor”, rather an extension of yourself, trained on your preferences, your writings, your history of decisions.

“Personal agents” and digital twins: promise… and pitfalls

The central idea: Personal LLMs can solve the “attention problem” by preparing the context, summarizing the issues, and even voting for you on routine crypto topics. If the point is sensitive or ambiguous, the agent asks you for the key elements, instead of drowning you in 40 pages.

This concept does not come out of nowhere: on the Near Foundation side, a researcher has already described work on “digital twins” capable of voting on behalf of members to counter low participation. Same direction, different laboratory: make crypto governance practicable at scale, without transforming it into an oligarchy.

The promise is attractive, but the pitfalls are very real: who controls the AI, how to prevent it from being manipulated, and above all how to ensure that it actually votes without exposing your privacy?

For the idea to stand up, solid safeguards are needed: clear explanations for each voting decision, confidence thresholds, a human veto, and a layer of cryptography to protect the user.

Vitalik also defends this logic of sovereignty at the protocol level. We see this with Ethereum, which is preparing an update designed to be more resistant to censorship. Otherwise, we do not reduce centralization: we simply move it from delegates to an elite of models and operators.