Vitalik Buterin Unveiled System for Private AI Agent Operations

Published: October 17, 2026 – 08:17 UTC

Vitalik Buterin is testing private payments for AI agents.
To do this, he combined Tor and zkAPI.
The system isolates conversations via Tor and rewrites AI prompts locally.

Ethereum co-founder Vitalik Buterin demonstrated an experimental setup to improve AI agent privacy that combines the zkAPI protocol, the Tor network, cover traffic, and local prompt rewriting using a language model.

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 ensures you don’t leak your identity through payments, and all prompts are locally rewritten (bring-your-own-local-model; I’m using qwen3.8-flash-next)https://t.co/MepCFCrLv1My agent harness is wrapped with a bwrap sandbox (also experimenting with MicroVM) and wrapped again with torsocks, so everything it does goes through Tor.Here’s what me asking it to do a simple research task looks like.It’s getting quite usable! Though you definitely still pay quite a bit in increased latency, especially when starting a new conversation.Also, Tor is far from optimal here for privacy; the cover traffic mechanism is okay against passive adversaries but much weaker if the LLM provider is colluding with the guard and actively manipulating timing. This kind of task really requires a more specialized mixnet, probably take the tradeoff of a higher bandwidth multiplier and built-in constant cover traffic.Also also, much cheaper zkAPI deposit and withdraw gas costs are coming to mainnet soon.
— vitalik.eth (@VitalikButerin) October 16, 2026

The solution is integrated with his agent environment pi and is designed to hide the link between the user, their payments, and their requests to AI services.

How the Privacy System for AI Agents Works

According to Buterin, the system assigns each conversation a separate Tor circuit, adds random cover traffic, and uses zkAPI to avoid revealing the user’s identity through payments. Prompts are also rewritten locally using a language model — developers can choose their own model, and Buterin is using Qwen3.8-Flash-Next.

Recall that recently Ethereum Foundation researcher Justin Drake and Buterin urged the crypto industry to prepare for AI-driven private key compromises. The Incrypted editorial team collected experts’ views on how likely this is:

The agent environment is additionally launched in an isolated sandbox using bubblewrap, and all of its network traffic is routed through Tor using torsocks. Buterin is also experimenting with using micro virtual machines (MicroVMs) for isolation.

“It’s getting quite usable!” he noted, while adding that the system significantly increases latency, especially when starting a new conversation.

The developer emphasized that Tor is not an optimal solution for this level of privacy. Cover traffic can help against passive surveillance, but its effectiveness is much lower if the language model provider cooperates with a Tor entry node and deliberately manipulates data transmission timing.

In Buterin’s view, these tasks require a specialized traffic-mixing network (mixnet) that would provide a constant cover stream at the cost of higher bandwidth consumption.

zkAPI Hides the Link Between Payments and Requests

zkAPI is a protocol developed by Open Anonymity in collaboration with the Ethereum Foundation. It enables users to pay for access to AI and other APIs in Ethereum (ETH) without revealing which onchain deposit the funds come from for a given service.

A user deposits Ethereum (ETH), pays for actual service usage, and can withdraw any unspent balance onchain. Thanks to zero-knowledge proofs, the service verifies that funds are available for payment without receiving information about the user’s specific deposit.

The project has the OA Chat web interface, as well as a local client compatible with the OpenAI API, which can be configured to work with Open WebUI and other compatible apps.

However, the configuration described remains experimental. In the development branch referenced by Buterin, the production proving keys are not yet ready, and the required network configurations for the main protocol have not been deployed. Therefore, the relevant privacy features should not be considered a production-ready solution.

Buterin also said that cheaper deposit and withdrawal operations via zkAPI should appear on the Ethereum mainnet in the near future.

Earlier, we wrote that Buterin assessed scenarios for the emergence of artificial superintelligence by 2040.

Сообщение Vitalik Buterin Unveiled System for Private AI Agent Operations появились сначала на INCRYPTED.

Source: Incrypted