Ethereum co-founder Vitalik Buterin unveiled a proposal for a high-level programming language designed to improve how humans read AI-generated mathematical proofs. This language would compile directly into formal proof assistants like Lean or HOL, which are tools used to verify correctness step by step.
Bridging the Gap Between AI Proofs and Human Readers
AI increasingly automates the creation of extensive formal proofs, often outpacing human ability to produce similar work manually. However, these proofs' internal steps focus solely on mathematical correctness and are rarely inspected directly by readers. Instead, understanding relies on clear definitions and theorems, which convey the guarantees software provides. Buterin emphasized the need for a language optimized for presenting these readable components rather than the proofs themselves.
Lean, a proof assistant with a six-decade history, is actively employed by Ethereum researchers to verify cryptographic algorithms and consensus mechanisms. Buterin's proposal complements Ethereum's broader rebuilding efforts, including the Lean Ethereum roadmap and the ongoing development of a formally verified zero-knowledge Ethereum Virtual Machine (ZK-EVM). These initiatives rely on similar formal methods to increase security and reliability.
Large language models like Claude and Deepseek 4 Pro can already generate usable proofs in Lean, demonstrating AI's growing role in formal verification. Projects such as evm-asm have produced EVM implementations formally verified against readable references. This capability was highlighted by participants in Buterin's recent AI challenge, where testers completed verification tasks within hours.
Security pressures are mounting as AI-assisted exploit attempts rise, making formally verified code an essential defense. A more accessible specification language would enable developers to audit software claims quickly without parsing extensive proof details.
This material is informational and does not constitute financial advice.
