Vitalik Buterin Creates ‘Paranoid Mode’ to Protect AI Agent Privacy

An AI assistant helping manage investments may expose your portfolio, trading strategies, and financial questions to the firm operating its model. Ethereum co-founder Vitalik Buterin has devised a method to make that data trail harder to follow.

On October 16, Buterin unveiled a "paranoid mode" for his AI agent, designed to hide user identity, request sources, and payment methods. The code is publicly available, but the combined system is still experimental.

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…

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

Financial Privacy Risks from AI Agents

As AI agents gain access to wallets and financial tools, their conversations may reveal sensitive financial information. A trading assistant could disclose which tokens it monitors or details about a user's portfolio.

Even users trying to stay anonymous might be identified through payment records, IP addresses, or repeated details in prompts.

Buterin addresses those clues in four steps.

Tor gives each conversation a separate route to hide the user's IP address. Bogus requests disguise real traffic. A small AI model on the user's computer rewrites prompts to remove identifying details.

Finally, zkAPI uses cryptographic proofs to show that requests are paid for without identifying the payer.

The payment system already runs on Ethereum mainnet, accepting deposits in assets such as ETH and USDC. Buterin also limits his agent to a software sandbox to restrict its access to his computer.

Impplications for Investors

Buterin says the system is "getting quite usable," although each new conversation starts more slowly.

Full anonymity still not possible. Providers still see rewritten prompts, which could contain recognizable information.

He acknowledged another weakness: an AI provider cooperating with Tor's first relay could examine traffic timing and potentially identify real requests. He prefers a more specialized privacy network, despite higher data costs.

definitely no point trying to de-link within a session – this daemon treats a session as a unit, and gains efficiency from *not* refreshing the Tor circuit or doing a zkAPI payment for each message within a session

each new session gets its own Tor circuit + zkAPI payment though

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

For investors, the longer-term question is whether private AI payments become cheap and fast enough for trading tools and other financial services.

Buterin says lower zkAPI transaction fees are coming. His prototype suggests a potential use for Ethereum infrastructure, but offers no evidence of widespread adoption or an immediate benefit to ETH's price.