Vitalik Buterin unveils privacy-focused AI agent system

Ethereum co-founder Vitalik Buterin has introduced an experimental AI agent system designed to minimize identity exposure and data leakage when using external AI services. The system integrates several technologies to protect users across three main vectors: payment information, network connectivity, and prompt content.

To protect network anonymity, the system assigns a separate Tor circuit to each conversation and utilizes random fake traffic to obscure communication patterns. To prevent identity disclosure through prompt content, the agent uses a local Large Language Model (LLM)—specifically mentioning the Qwen model—to rewrite prompts before they are sent to external providers. This process aims to strip away linguistic patterns or specific identifiers that could link a user to their queries.

For financial privacy, the system employs zkAPI, which uses zero-knowledge proofs to verify that a valid payment exists without directly linking the user's identity to their specific API usage history.

Buterin noted that while the tool is becoming functional, it currently suffers from significant latency, particularly when initiating new conversations. He also highlighted technical limitations, noting that Tor remains vulnerable to correlation attacks if service providers and entry nodes collude, suggesting that a dedicated mix network might be required for more robust protection in the future.