Vitalik's new DAO governance proposal uses personal LLMs, ZK proofs, and prediction markets to fix participation without handing power to machines.
DAO governance has had a quiet, persistent problem for years that nobody in crypto really wants to say out loud: most people don't vote, and the ones who do are often just following the largest token holders. Delegation was supposed to fix this. It didn't. It just moved the power concentration one step further from view.
Vitalik Buterin posted his most detailed governance proposal yet today, and what's interesting about it is that he doesn't frame the problem as apathy. He frames it as attention. There are thousands of decisions to make across many domains of expertise, and most people lack the time or skill to be experts in even one, let alone all of them. That's not laziness — that's reality. And the conventional solution of clicking a delegate button means you hand over your voice entirely the moment you make that choice.
His proposed answer is a personal AI agent — specifically a personal LLM trained on your writing, your conversations, your stated values — that votes on your behalf across the hundreds of routine governance decisions that come up in any active DAO. When the agent is unsure how you'd decide on something, it pauses and asks. The rest it handles. You stay connected to the process without being consumed by it.
But the piece that stood out to me more than the voting mechanic is what he's proposing on the privacy side. AI agents would operate within MPC environments or trusted execution environments, enabling them to process private data without leaking it to the public blockchain. And zero-knowledge proofs would allow users to prove they're eligible to vote without revealing their wallet address or how they voted — guarding against coercion, bribery, and whale watching.
That last part matters. One of the subtler failure modes in token governance is that smaller holders often just copy the votes of large wallets once they see them. ZK-based anonymity breaks that dynamic structurally.
The prediction market layer is a clever addition too. As generative AI floods open governance forums with low-effort proposals, signal-to-noise becomes a serious problem. Buterin's suggestion — let agents bet on whether proposals will pass, rewarding accurate bets — creates a financial incentive to surface quality and bury noise without requiring a human committee to make those calls.
The broader framing aligns with what Buterin has called "defensive acceleration" — using technology to strengthen decentralized cooperation and societal resilience, specifically while avoiding concentration of power. That's the through-line in almost everything he's written about AI and Ethereum lately. Not acceleration for its own sake. Not AI replacing human judgment. AI scaled to the volume of decisions humans were never equipped to handle alone.
Whether DAOs actually implement any of this is the harder question. The cryptographic infrastructure — ZK proofs baked into every governance tool, MPC environments for sensitive decisions — is expensive and slow at scale right now. But the design logic is tight. And Vitalik has a way of proposing things that look theoretical until they don't.
