Ethereum Founder Vitalik Buterin Says AI Won’t Doom Crypto Security

Vitalik Buterin pushed back against the view that artificial intelligence will inevitably doom cybersecurity, arguing instead that AI could shift the advantage toward defenders. In a post on X, the Ethereum co-founder suggested that sufficiently powerful models might make formal verification — the mathematical proof that a program meets specified properties — practical even for large, complicated systems. He framed the challenge as less about whether AI can break systems and more about establishing precise, usable definitions of what 'secure' means. Buterin pointed to recent activity in the crypto ecosystem where AI has been used both offensively and defensively. Researchers and security teams have used AI agents to scan code, test exploits, and identify bugs before attackers could exploit them; at the same time, attackers have leveraged AI to find flaws. High-profile incidents cited include a Zcash Orchard vulnerability found by security researcher Taylor Hornby using Anthropic’s Claude Opus 4.8, and a Coldcard firmware weakness that attackers exploited to steal roughly $130 million in Bitcoin. The acceleration of AI-driven discovery has prompted projects across the space to incorporate AI into their security toolkits. In July and August, multiple projects reported AI-identified issues in critical infrastructure and Bitcoin software, prompting some services to pause operations while auditors investigated. Volunteer teams have also adopted AI-assisted auditing at scale: the Bitcoin Red Team reported flagging thousands of potential vulnerabilities across hundreds of projects. Buterin said Ethereum intends to pursue whole-program verification and related AI-assisted security measures over the coming years, calling it essential for blockchains aiming for scalability and privacy. He acknowledged that proving a program is secure is only part of the task — defining security requirements clearly remains the harder problem — but emphasized substantial progress has already been made and that AI could make rigorous proofs feasible for real-world systems.