In April 2026, Ethereum co-founder Vitalik Buterin transitioned from cloud AI, opting for a secure, local setup using an Nvidia 5090 GPU that processes 90 tokens per second with the Qwen3.5:35B model. Highlighting security concerns, he noted that about 15% of AI agent skills harbor malicious directives, as reported by Hiddenlayer. To combat these vulnerabilities, Buterin developed a “self-sovereign” AI system, enforcing a “human-plus-LLM 2-of-2” confirmation rule for all outgoing communications. He emphasized the importance of sandboxing for security and implemented tools like bubblewrap to isolate processes. Moreover, he open-sourced a messaging daemon that requires human verification for third-party messages, enhancing security. On Ethereum integration, he warned against granting AI agents unmonitored wallet access and proposed privacy-preserving methods for remote inference, such as ZK-API and mixnets. Buterin’s insights encourage a cautious approach to AI deployment, highlighting privacy as paramount in an evolving digital landscape.
Source link
