Vitalik Buterin: AI And Formal Verification Can Make Critical Code Unhackable

by
Alisa Davidson

Published: May 18, 2026 at 10:32 am Updated: May 18, 2026 at 10:32 am

by Anastasiia O

Edited and fact-checked:
May 18, 2026 at 10:32 am

To improve your local-language experience, sometimes we employ an auto-translation plugin. Please note auto-translation may not be accurate, so read original article for precise information.

Ethereum co-founder Vitalik Buterin has never been shy about big ideas. But his latest essay ventures beyond blockchain into something that could reshape the foundations of software security itself: formal verification, the practice of writing mathematically checkable proofs that computer code behaves exactly as intended. In the face of increasingly powerful AI that can find and exploit software bugs at scale, Buterin argues this centuries-old approach to mathematical proof is not just useful — it may be the only credible path to a trustworthy digital future.

The argument is timely. AI-assisted bug discovery is rapidly tilting the playing field toward attackers. Code that once took teams of human auditors weeks to review can now be scanned for vulnerabilities in minutes. Some voices in the security world have responded to this reality with resignation, suggesting that deterministic software guarantees are essentially over, or that the only viable response is retreating behind closed-source walls. Buterin firmly rejects both conclusions.

His optimism is grounded not in wishful thinking, but in a specific technological pairing: AI and formal verification used together. AI can generate enormous volumes of code, including highly optimized low-level assembly that would be painstaking for humans to write. Formal verification can then prove, with machine-checkable mathematical certainty, that this code has the desired properties. The result, Buterin suggests, is a return to writing maximally efficient code — the kind programmers wrote fifty years ago in raw assembly — but this time with a rigorous proof of correctness attached. Researcher Yoichi Hirai calls this “the final form of software development.” Buterin is inclined to agree.

What Formal Verification Actually Does — and Doesn’t Do

To understand Buterin’s case, it helps to be clear about what formal verification is. In essence, it means writing mathematical theorems about your software and then checking those theorems automatically. Rather than testing whether code works on a sample of inputs, you prove that it works on all possible inputs, given certain defined assumptions. The Lean programming language, increasingly used in both pure mathematics and software engineering, is the primary tool here. Projects already underway include formally verified implementations of cryptographic protocols like Signal’s X3DH key exchange, ZK-STARK proof systems, and even a full EVM (Ethereum Virtual Machine) built directly in RISC-V assembly with proofs of correctness attached.

This is genuinely powerful. The nastiest software bugs are often interaction bugs — flaws that sit at the boundary between two subsystems that were each considered sound in isolation. Human auditors simply cannot hold an entire complex system in mind simultaneously. An automated proof-checking system can. Formal verification is also uniquely well-suited to the kinds of systems Ethereum most needs to get right: quantum-resistant signatures, zero-knowledge proof systems, and consensus algorithms — all areas where the security properties are conceptually simple to state, even if the implementations are fiendishly complex to build.

But Buterin is careful not to oversell it. “Provable correctness” does not mean what most people assume it means. A proof only demonstrates that the code satisfies a formally stated specification. If the specification is incomplete, the proof is incomplete. If critical assumptions baked into the proof don’t hold in practice — say, a hardware side-channel leaks information in ways the threat model never considered — the proof is still valid but the system is still insecure. History offers sobering examples: formally verified C compilers have shipped with bugs; formally proven cryptographic protocols have later been broken under adversary models their authors didn’t anticipate. Formal verification, Buterin stresses, is not a silver bullet. It is one powerful technique among several, and it fails when applied carelessly, partially, or with a specification that doesn’t match what users actually need.

The Road Ahead: A Secure Core in a Buggy World

Where Buterin lands is a nuanced but genuinely hopeful vision. The future of software security, in his framing, is not a world where all code is perfectly verified — that is neither achievable nor necessary. It is a world split between a hardened, shrinking “secure core” and a looser, sandboxed periphery. The peripheral code — apps, plugins, AI-generated scripts — may remain messy and bug-prone. That’s acceptable, so long as it runs with minimal privileges and cannot compromise the core. The secure core, by contrast — operating system kernels, Ethereum itself, cryptographic infrastructure, IoT foundations — must be held to an entirely different standard, and formal verification is central to meeting it.

In this architecture, AI changes the equation not by making code safer by default, but by making rigorous verification tractable for the first time. Writing proofs by hand is notoriously difficult and has kept formal methods a niche discipline for decades. But if AI can write both the optimized implementation and the accompanying proof, with human oversight focused narrowly on checking that the stated theorems actually capture what matters, the calculus shifts. The hard work of verification becomes automatable; the human role becomes one of specification and judgment rather than line-by-line drudgery.

The stakes, in Buterin’s view, go beyond Ethereum or even cryptocurrency. The cypherpunk tradition — the belief that on a digital network, defenders have a structural advantage because building cryptographic protection is easier than breaking it — is in genuine danger from AI-powered attackers. Formal verification, combined with AI, is one of the few available tools capable of restoring that advantage. Not by eliminating all bugs everywhere, but by making the most critical systems genuinely provably secure against a formally defined class of threats. In a world of increasingly autonomous, increasingly capable AI, that may be precisely the kind of hard guarantee we need.

Alisa, a dedicated journalist at the MPost, specializes in crypto, AI, investments, and the expansive realm of Web3. With a keen eye for emerging trends and technologies, she delivers comprehensive coverage to inform and engage readers in the ever-evolving landscape of digital finance.

More articles