On October 16, Ethereum co-founder Vitalik Buterin introduced a privacy-focused 'paranoid mode' for his AI agent. The experimental setup aims to obscure user identity, request origins, and payment details. The code is currently available to the public.
The system employs four primary mechanisms to enhance privacy. It uses Tor to assign a unique circuit to each conversation, masking IP addresses. Fake traffic is generated to disguise actual requests, while a local LLM rewrites prompts to remove identifying information. Finally, zkAPI utilizes cryptographic proofs to verify payments without revealing the payer's identity.
The payment infrastructure operates on the Ethereum mainnet and supports assets including ETH and USDC. Buterin noted that while the system is becoming usable, it faces limitations, such as increased latency at the start of new sessions. He also acknowledged that traffic timing analysis could potentially deanonymize requests.
Buterin is exploring the use of specialized privacy networks to improve security, despite higher data costs. He also indicated that lower transaction fees for zkAPI are expected in the future. The project demonstrates a potential application for Ethereum infrastructure in AI, though it remains in an experimental phase.
