Vitalik Buterin laid out a practical view of how AI and Ethereum could work together, centering his comments on privacy, verification, on-chain agent activity, and the direction of Layer 2 development. He said new technologies are needed so users can interact with large language models without exposing their identities or personal data.
Local AI and zero-knowledge tools sit at the core of privacy
On privacy, Buterin pointed to several approaches: AI software running locally on personal devices, zero-knowledge proofs enabling anonymous API access, and stronger cryptographic verification methods. The aim is straightforward. Users should be able to use LLMs without handing over unnecessary personal information.
He also described a role for AI-based intermediary teams that could stand between users and the blockchain. These agents could audit transactions, flag suspicious activity, and suggest actions before users make on-chain decisions. In his framing, that model could put the idea of “trust, but verify” into day-to-day use.
AI agents could transact, post guarantees, and act for users
Buterin’s comments also covered the prospect of AI agents taking part in economic activity on Ethereum. He said such agents could conduct transactions, make API calls, deposit security guarantees, and operate on-chain on behalf of users. If that model develops, access to crypto systems could become easier while execution becomes more efficient.
He linked that possibility to decentralized governance as well. In areas where human attention and decision capacity are limited, including decentralized governance and prediction markets, LLMs could help collective decision-making process more information. Buterin described this as “scaling human judgment,” with Ethereum’s community-centered systems potentially improving how they process information.
Buterin urges Layer 2 builders to pursue new ideas
Beyond AI, Buterin called on developers in Ethereum’s Layer 2 ecosystem to put innovation first. He said Ethereum’s next phase of progress should come from new ideas rather than repeating applications that already exist.
His remarks place AI and blockchain on a shared track: solve privacy and verification problems, let AI agents participate in on-chain economic activity, and push L2 teams toward original design choices. The source material did not include a product timeline or named implementations, but it made his priorities clear.