Ethereum co-founder Vitalik Buterin argues that AI-assisted formal verification could become essential for protecting blockchain networks from sophisticated cyberattacks and software vulnerabilities. In a blog post Monday, Buterin outlined how this technique-using mathematical proofs to confirm code behaves correctly-could reduce risks in smart contracts, cryptographic systems, and core Ethereum infrastructure.
Buterin wrote that formal verification can prove not just that a protocol is secure in theory, but that the actual code users run is secure in practice. He emphasized that this "end-to-end" approach greatly improves trustlessness: users only need to check proven statements about the code, not the entire codebase.
Buterin's comments come as advanced AI models like Anthropic's Claude Mythos demonstrate alarming offensive capabilities. In April, Claude Mythos found 271 vulnerabilities in Mozilla Firefox during internal testing, and earlier this month researchers used a preview version to develop an exploit targeting Apple's M5 chip protections. OpenAI's GPT-5.5 has similarly shown advanced cyberattack abilities, raising concerns among governments and security agencies.
For crypto projects, software flaws can be catastrophic. In April, North Korea-backed Lazarus Group stole $292 million from Kelp DAO's infrastructure. North Korean state-sponsored hackers have stolen over $6 billion in cryptocurrency to date, underscoring the urgency of better defenses.
Buterin cautioned that formal verification is not a panacea, but said it is particularly well-suited for complex technology like quantum-resistant signatures, STARKs, consensus algorithms, and ZK-EVMs-critical components for Ethereum's next major iteration. He rejected the idea that AI will ultimately make open-source systems impossible to secure, arguing that the defender still holds the advantage in cyberspace. Future systems, he said, will likely rely on a small, highly secured "core" infrastructure protected through formal verification and restricted security environments.