Ethereum cofounder Vitalik Buterin has trialed a multilayered strategy to secure personal data while requesting AI-generated health recommendations. Drawing on his own health and travel information, Buterin aimed to access tailored diet and exercise advice without exposing sensitive details to external services.
Local coordination with privacy layers
Buterin’s experiment involved a local AI model, Qwen 3.8 Flash Next, which routed requests to advanced remote models as needed. A key component of his approach was minimizing the amount of personal data shared during each query.
Three privacy layers underpinned his setup. The first focused on prompt security: the local model not only composed and filtered questions, but also masked identifying language or stylistic clues, lowering the risk of reidentification by outside models.
For transaction privacy, Buterin utilized zkAPI, a system that separates service payments from user identity. Described by the Ethereum Foundation as a private usage credit protocol, zkAPI enables users to load funds into a vault and authorize spending through zero-knowledge proofs rather than linking payments to specific identities or requests.
In practice, temporary API keys managed spending caps while signed usage receipts determined exact charges. The payment provider could confirm funding through proofs, yet never accessed the content of AI queries submitted directly to remote models.
The third layer employed Tor routing. By running the zkAPI client over Tor, Buterin prevented network-based identification, concealing IP addresses and other network data from service providers.
Each layer in this system addresses a distinct privacy risk. Even if names are omitted from AI prompts, linking payment records or network footprints can still reveal the requester’s identity, meaning a multi-pronged approach is necessary for robust protection.
According to documentation shared by the Foundation, repeated inclusion of personal details, history reuse, or consistent writing style can still permit models to link different sessions to the same individual, signaling the inherent complexity of privacy in AI interactions.
Performance and tradeoffs
Buterin acknowledged improvements in the health advice generated through this approach, with richer responses traced to the advanced capabilities of remote models. However, his experiment exposed several limitations affecting both speed and privacy.
Latency emerged as a significant challenge, as routing through Tor slowed response times by a factor of 10 to 100 compared to direct connections. He observed that Tor’s network separation could be insufficient for anonymizing requests, reducing its effectiveness as a privacy layer in this context.
On the local machine, Qwen managed to process only 20 to 30 tokens per second, far short of Buterin’s target of over 100 tokens per second for seamless usage. He also stressed the critical role of carefully crafted prompt composition, noting that insufficient or excessive redaction could limit both privacy and the quality of AI recommendations.
Adjusting information sharing proved complex: holding back too much context weakened the practical value of responses, while detailed prompts increased exposure risk. This tradeoff points to the necessity for continuous refinement in privacy-first AI workflows.
Context: Toward cryptographic world computing
This AI experiment builds on ideas Buterin outlined in a recent essay, where he discussed Ethereum’s transition toward a cryptographic world computer model. The foundation for this vision includes enhanced privacy, formal verification, and decentralized off-chain computation powered by advanced cryptography.
Hegotá, identified as a likely final conventional Ethereum fork before the introduction of quantum-resistant methods and recursive zero-knowledge proofs, marks a major step in these ongoing developments.
Given the volatility and rapid shifts in crypto markets, where a single Fed decision or a sudden altcoin listing can quickly reshape portfolios, traders increasingly rely on integrated solutions for monitoring. Smart investors are adopting privacy-focused platforms like CryptoAppsy, which combine real-time charts, detailed coin data, portfolio tracking, and macroeconomic alerts—all without registration—helping users efficiently manage their portfolios in a unified interface.
