Ethereum co-founder Vitalik Buterin has uploaded a branch with a paranoid mode for AI agents on GitHub. The system assigns each conversation a separate Tor channel, applies zkAPI to protect payment privacy, and rewrites prompts locally using the Qwen model. According to the repository, zkAPI allows for paying AI and other APIs in ETH and sending requests without disclosing asset holder information. The project was created by Open Anonymity in collaboration with the Ethereum Foundation.
Buterin's personal agent uses the bwrap sandbox, sometimes MicroVM, and an additional torsocks shell. The developer called the build usable but warned that privacy comes at the cost of latency, especially when starting a new dialogue. He considers Tor not the best solution for such a model. Traffic masking is acceptable against a passive observer but significantly weaker if the LLM provider colludes with a Tor node and actively manipulates timing. In his assessment, a specialized mixnet with constant cover traffic would be more suitable for such scenarios.
Buterin also wrote that cheaper deposit and withdrawal fees in zkAPI should soon appear on the mainnet. Recall that in September, the programmer described transforming Ethereum into a cryptographic world computer, a system that combines blockchain, privacy, verification, and decentralized off-chain components.
The publication shows how the developer approaches privacy protection at the individual user level. Instead of relying on a single tool, he builds multiple layers: process isolation, separate communication channels, and payments that do not disclose the owner. This approach requires carefulness and a willingness to accept trade-offs– primarily in terms of speed. The branch on GitHub is open for review, allowing the community to evaluate the solutions and offer their own.
This material is prepared solely for informational purposes and does not constitute financial advice or a recommendation.
Related articles
