BTC
ETH
SOL
XRP
BNB
TON
DOT
LTC
POL
HBAR
NEAR
ARB
UNIVitalik Buterin says his 'paranoid mode' AI agent hides user identity
The Ethereum co-founder has combined anonymous crypto payments, Tor routing and a locally run model to stop AI services linking prompts to users, though he says latency remains high.
Why it matters: The experiment shows how blockchain payments, privacy networks and local models could be combined to limit what AI providers learn about users, a growing overlap between crypto infrastructure and AI agents.
Key points
- Buterin's Pi agent harness pairs the Ethereum-based zkAPI payment system, Tor, random cover traffic and a locally running LLM.
- Each conversation gets its own Tor circuit, and a local model rewrites prompts before they are sent to external providers.
- zkAPI, launched 1 October by the Ethereum Foundation and Open Anonymity Project, uses zero-knowledge proofs to separate payment identities from AI requests.
Vitalik Buterin says his 'paranoid mode' AI agent setup is designed to keep AI services from identifying the people who use them.
The setup targets the ways AI providers can single out users: payment records, IP addresses, recognisable writing patterns and the contents of prompts. Each conversation is given a separate Tor circuit, while extra randomised traffic hides genuine requests. A local model rewrites prompts before they go to outside services, cutting the chance that identifying details reach the provider. Buterin said he used Qwen3.8-Flash-Next for that local processing, though other models can be substituted, and the features sit in an OpenAI-compatible daemon.
He added further isolation by running the Pi harness inside a Bubblewrap sandbox and routing its connections through Tor with torsocks, and is testing micro-virtual machines as another layer. The zkAPI component, launched on 1 October by the Ethereum Foundation and Open Anonymity Project, lets users deposit ETH or USDC into an Ethereum vault and receive private credits, with temporary API credentials and spending limits.
Buterin acknowledged the system is not perfect, saying it is 'getting quite usable' but that users pay in increased latency, especially when starting a new conversation.
Live updates on NewsFeeds:EthereumDesk editor: Keshav Aggarwal (accountable for this page; this summary was not edited by hand)
How this was produced: an AI model wrote this summary of Bitcoin.com News’s report, which we credit and link above. Read our Editorial and AI policy. Spotted a mistake? See Corrections.
Not investment advice. Read the risk disclaimer.