Vitalik: AI should not have free access to personal data; it needs to be localized, sandboxed, and subject to dual human-machine verification.

On April 2nd, CoinFeed reported that Ethereum co-founder Vitalik Buterin published an article detailing his exploration of localized, private, and secure solutions for personal use of AI. He pointed out that the current AI field (including local open-source AI) is extremely lax in terms of privacy and security. For example, OpenClaw agents can modify critical settings without human confirmation, malicious external input can easily take over user instances, and some skills contain malicious instructions. Vitalik advocates that all LLM inference and documentation should be prioritized locally, with everything sandboxed and isolated. He tested hardware such as an NVIDIA 5090 laptop and an AMD Ryzen AI Max Pro, using the Qwen3.5:35B model, running through llama-server, and using the NixOS system.