Ethereum’s Vitalik Buterin Builds Privacy-Focused AI Agent That Limits Identity and Data Leaks

Ethereum co-founder Vitalik Buterin has shared an experimental “paranoid mode” wrapper for his AI agent harness that layers Tor, zkAPI payments, cover traffic, and local prompt rewriting to limit identity and data exposure when using remote language models.

Ad

Ad

Vitalik Buterin Unveils Privacy-Focused AI Agent Using Tor and zkAPI

In an X post on October 16, Ethereum co-founder Vitalik Buterin revealed his personally developed privacy-focused AI Agent stack. He integrated zkAPI, Tor, random cover traffic, and local LLM prompt rewriting into the pi agent harness.

AI agent assigned an independent Tor circuit to each conversation, prevented payments-linked identity data leak through zkAPI, and hid requests with random cover traffic. He wrote all prompts using a Qwen3.8 model.

“My agent harness is wrapped with a bwrap sandbox (also experimenting with MicroVM) and wrapped again with torsocks, so everything it does goes through Tor,” he said.

Vitalik Buterin stated that the current solution already has a certain level of practicality, but starting new conversations still involves noticeable latency. He added that Tor’s protection against collusion between service providers and entry nodes for time-correlation analysis is limited.

In the future, a dedicated mixnet may be needed. He also revealed that the gas costs for deposits and withdrawals on zkAPI via the Ethereum mainnet are about to decrease further.

Ad

Ad

Ethereum Co-Founder Tests System and Shares Suggestions

Vitalik Buterin shared screenshots of the AI agent system handling a simple research task of identifying top vegan restaurants in Tallinn. He described the overall result as “getting quite usable.”

Wrapped zkAPI + Tor + extra cover traffic + local LLM prompt rewriting in an open-AI compatible daemon, and made a "paranoid mode" wrapper for my agent harness (pi):

Each conversation gets assigned a separate Tor circuit, your requests are masked with random cover traffic, zkAPI… pic.twitter.com/2y236zA0cw

— vitalik.eth (@VitalikButerin) October 16, 2026

However, he noted increased latency, especially when starting a new conversation. He also stated that Tor is not optimal for privacy. Cover traffic offers reasonable protection against passive observers, but it is weaker if an LLM provider colludes with a Tor guard node and actively manipulates timing.

He suggested a specialized mixnet with constant cover traffic and a higher bandwidth multiplier would better suit the problem.

The Github repository described zkAPI as a system that lets users pay for AI and other APIs with Ethereum while keeping requests unlinkable from their deposits, using zero-knowledge proofs.

Recently at TOKEN2049, Vitalik Buterin said AI agents will replace crypto wallets and front-end interfaces while blockchains coordinate interactions.

As CoinGape reported earlier, Ethereum co-founder warned AI could weaken lattice-based cryptography, including schemes considered quantum-resistant, within two years. He argued that this is the reason Ethereum’s lean roadmap has focused on a “hash-only” direction rather than lattices.

Also Check: Web3 AI Agents Directory That Combines LLM Reasoning with On-Chain Execution