Wu Shuo learned that Ethereum co-founder Vitalik Buterin posted that he is testing a privacy-focused AI architecture that uses personal health and travel data to obtain diet and exercise recommendations, while trying to minimize the leakage of private information to remote large models. The scheme is coordinated by the local model Qwen 3.8 Flash Next, and employs a three-layer mechanism to reduce identity exposure: the local model rewrites queries to conceal personal information and writing characteristics, zkAPI hides payment associations, and Tor hides network and IP information. Vitalik said the scheme is already operational, but Tor latency is still about 10–100 times higher. The local model operates at roughly 20–30 TPS, and the stricter the privacy protection, the less information the remote model can use.