Ethereum co-founder Vitalik Buterin has proposed a new programming language that would compile directly into Lean or HOL, another formal proof assistant. According to reports from CoinDesk, the language is specifically designed to address a gap in how people read AI output, as artificial intelligence increasingly produces large blocks of automated proofs that are often faster than human teams can write by hand. The proposal targets the challenge that few readers can quickly confirm what these proofs actually establish. Buterin has clarified that the goal is literally what he asked for: specs = as human readable as possible - proofs = AI written, can be ugly, who cares not sure how it could be any simpler!
Lean is a proof assistant that mathematicians and engineers use to write proofs that computers can check line by line. As reported by CoinDesk, Ethereum researchers already use Lean to verify cryptographic code and consensus logic. In his proposal, Buterin argues that a proof's internal steps carry only one requirement - mathematical correctness - and that readers never inspect that machinery directly. The timing aligns with Ethereum's own rebuild effort, known as the Lean Ethereum roadmap, where researchers are building a formally verified ZK-EVM using comparable methods.
According to the proposal, large language models can already write usable Lean proofs. Buterin has named Claude and Deepseek 4 Pro as capable tools, alongside Leanstral, a smaller model specifically tuned for Lean. One example project is evm-asm, an EVM implementation verified against a readable reference. The capability echoes the reasoning skills developers displayed in a recent Buterin AI challenge, where testers solved the challenge within hours. Buterin has confirmed that Bend2 is done, Fable is now emulating users so we catch bugs and improve the UI / UX before launch, indicating active development progress.
The proposal extends beyond convenience considerations, as security researchers have tracked a jump in AI-assisted exploit attempts this year. As reported by CoinDesk, formally verified code offers defense against this trend. A friendlier specification language would allow developers to audit claims without wading through surrounding proof. Buterin continues testing these ideas publicly, recently demonstrating an anonymous billboard built with zero-knowledge proofs to show how verifiable claims can move from research repositories into working products.
