Vitalik Buterin Tests Privacy-Preserving AI Recommendations Using Health and Travel Data

Vitalik Buterin has shared the results of an experiment that allows users to leverage personal health and travel data for tailored diet and exercise recommendations while preventing sensitive information from being exposed to external AI models, according to a report by blockchain media U.Today on October 4 (local time).
The goal was to provide personalized suggestions without revealing private details. Buterin used the local model Qwen 3.8 Flash Next as an orchestrator and connected remote models via tool calls only for complex reasoning or knowledge-intensive tasks.
Privacy protection was designed in three layers: The local model drafted queries sent to external models, reducing identity exposure through identifiable information and writing style; zkAPI was applied at the payment layer; and Tor was used at network and IP levels. Buterin emphasized all three were necessary.
The local model also utilized skill files to craft minimal information requests for remote models. zkAPI was deployed as a command-line tool wrapped in Tor. Buterin confirmed the recommendations were received, with external model data improving suggestions.
Limitations were noted: Tor is not optimized for request-level disconnection, potentially compromising privacy; latency increased 10-100x; skill file configurations need optimization; and Qwen 3.8 Flash Next felt slow to users (operating at 20-30 TPS versus the desired 100+ TPS for perceived speed).
Buterin also warned that restricting data provided to remote models too strictly would reduce their helpfulness.
In a separate post, Buterin reiterated Ethereum's direction, stating it will evolve into a cryptographic world computer by late September. He also suggested the upcoming Pectra hard fork in 2024 could be the last general-purpose fork visible from a technical perspective since 2015. After Pectra, recursive STARKs, automated formal verification, highly optimized consensus algorithms, and quantum-resistant technology will become key focuses. Buterin added that Ethereum is transitioning from a simple blockchain to a more robust system following PeerDAS, with this shift becoming the central narrative post-Pectra.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 부테린, 건강·여행 데이터로 개인정보 보호형 AI 추천 실험