Ethereum co-founder Vitalik Buterin said at the OKX Now 2026 summit in Singapore that AI is increasingly able to break out of sandboxes, attack websites, and discover software vulnerabilities. At the same time, he said, it is helping the Ethereum protocol find bugs, implement complex cryptography, and perform formal verification.
Stronger security may become both possible and required
Buterin said higher levels of security assurance will become feasible, and also necessary, because AI will keep finding weaknesses in systems that can be exploited.
AI may turn into a new user interface for blockchain activity
He said that about a month ago, he updated ENS information for the first time without using a user interface. Instead, he asked a local AI agent to write the script, and the task was completed in about five minutes.
Buterin said he expects direct execution of complex onchain operations through AI to gradually become normal, with AI serving as a new user interface in more scenarios.
Less reliance on web pages could cut one risk, but add others
Using the Safe-related hack that involved about $1.4 billion in losses as an example, Buterin said the attack targeted the interaction layer between users and onchain systems, rather than the onchain contract itself.
He said AI handling transactions directly could reduce risks created by reliance on web interfaces. Still, it would also introduce new challenges, including whether AI itself is secure, whether it can resist prompt injection, and whether it correctly understands onchain actions. As interaction methods change quickly, some issues may be resolved, while new security problems are also likely to emerge.