Vitalik Buterin (비탈릭 부테린), founder of Ethereum (ETH), presented the Chinese AI model DeepSeek V4 as a core technology for strengthening privacy on Ethereum. With an era approaching in which AI directly handles crypto wallets and blockchain interactions, he highlighted concerns that existing centralised AI structures make it hard to guarantee security and privacy.

On May 28, local time, blockchain outlet Cryptopolitan reported that Buterin outlined an approach combining local AI models with an Ethereum access layer. He said DeepSeek V4's local runtime environment is important for making private blockchain usage practical.

Buterin previously introduced the concept of "CROPS AI" at the ETH Mumbai conference in March. CROPS AI refers to AI with censorship resistance, open source, privacy and security. He warned at the time that AI could become a new security risk for the crypto ecosystem.

Buterin particularly noted that local AI is not automatically secure. He said that even as open-source AI frameworks and local agent environments spread, when difficult tasks arise they often end up calling OpenAI or Anthropic APIs. That means what appears to be local AI still relies on centralised servers in practice.

In that context, Buterin mentioned DeepSeek V4 as a realistic alternative. He assessed that a 2-bit quantised version of DeepSeek V4 can run in an environment with about 90 GB of memory and could play an important role in implementing a private local transaction-processing structure.

The structure Buterin envisions combines local AI with Ethereum zero-knowledge proofs (ZK). He said that when a user queries Ethereum data with a local model such as DeepSeek V4, the user can interact with the blockchain without exposing metadata, IP addresses or wallet balances to centralised RPC providers.

He said, "A zero-knowledge approach that sends paid calls to a remote large language model (LLM) can also be used to solve Ethereum's private RPC read problem." He added that combining private local LLM calls with zero-knowledge-based payments can handle blockchain interactions privately off-chain and help hide links between transactions.

Buterin stressed that "DeepSeek V4 is an example showing this kind of approach is possible on current hardware, not years from now." He said, however, that high-end hardware specifications are needed. He mentioned that Macs would need 96 GB to 128 GB of unified memory, while PCs would need a similar level of VRAM.

He also cited a DeepSeek V4 flash-optimisation patch for AMD environments as a future improvement task and predicted that the integration of AI and crypto infrastructure will expand further.

Buterin linked this trend to a "cypherpunk revival." He said that if AI takes on the role of a user's digital custodian, it becomes possible to separate payment information from user identity and even anonymise remote AI computation. He criticised the current open-source AI ecosystem for developing around performance without sufficiently considering privacy issues.

He predicted that the Ethereum ecosystem will need more AI models specialised for Ethereum. He said that strengthening smart contract and protocol code security requires models optimised for the Ethereum environment, not general-purpose AI. As a result, discussion of Ethereum privacy is moving beyond simple wallets and RPC infrastructure and expanding toward local AI runtime environments and the design of zero-knowledge-based access layers.