Vitalik Buterin argues AI-powered hacking threats are overstated and that modern AI can strengthen security through better verification and more human-readable security definitions.
AI & Agents ·
Vitalik Buterin has pushed back against concerns that AI-powered hacking will undermine cybersecurity, arguing instead that the right systems and stronger verification tools can tip the balance toward protection. He suggested that if AI can prove mathematical theorems, it could verify that "this program is secure" as a mathematical statement, even for complex code. However, he emphasized that defining what "secure" actually means is far harder—security concerns span multiple vectors including attacker interference, information leakage, and hardware or software failure—making human-readable definitions critical to practical verification.
Buterin sees advantages in verification through explicit definitions rather than scanning entire codebases. When definitions are clear, they become additive: two different security requirements can both be proven satisfied by the same program, or conflicts can be isolated as actionable problems. This contrasts with full-program verification, where a bug in any component can compromise the whole system. He identified message-passing protocols, sandboxes, and cryptographic systems like SNARKs as domains where definitions are meaningfully smaller than their implementations.
The approach does not apply equally to all software—user-interface components, for instance, may have definitions nearly as large as their code. Yet Buterin contends that modern AI changes the calculus by making verification more practical, shifting strategy away from hoping defenders find flaws first toward building inherently more resilient systems. He views this work as essential for blockchains pursuing scalability and privacy.