Ethereum co-founder Vitalik Buterin proposed a new programming language. It will be compiled directly into Lean or HOL, which is another formal proof-of-concept helper.
The idea targets a specific gap in how people read AI output. AI is increasingly producing large blocks of automated proofs, often faster than any human team can write them by hand. Few readers can quickly confirm what these proofs prove.
Language designed only for readers of AI-based guides
Lean is a proof plugin, a program used by mathematicians and engineers to write proofs that a computer can verify line by line. Ethereum researchers are already using it to verify cryptographic code and consensus logic. Proof assistants have been around for nearly 60 years, but the field has remained a niche field.
In his country mailButerin argued that the internal steps of proof carry only one requirement. This requirement is athletic health, nothing more. Readers never examine that machine directly. Definitions and theories work differently, as humans read those parts to find out what a program actually guarantees.
Buterin discovered a related dichotomy in May Blog mail. There, the mathematical proof shows that efficient low-level code conforms to a separate, readable specification, so a single audit covers both versions simultaneously.
Its timing also aligns with the rebuilding efforts of Ethereum, which has a separate moniker, Lean Ethereum Roadmap. Meanwhile, researchers are building a ZK-EVM has been officially verifieda provable, zero-knowledge version of the Ethereum Virtual Machine (EVM), using comparable methods.
AI writes the evidence, and humans verify the claims
Large language models can actually write usable Lean proofs. Buterin named Claude and Deepseek 4 Pro as capable tools, along with Leanstral, a smaller model tuned specifically for Lean. One example project is evm-asm, an EVM implementation that is checked against a readable reference. This ability echoes the logical thinking skills that Buterin’s AI developers have recently demonstrated challenge. Testers solved this challenge within hours.
However, the risks extend beyond comfort. Security researchers have tracked a jump in the use of artificial intelligence Exploitation attempts this year. Formally verified code offers one defense against this trend. The friendlier specification language will allow developers to review claims without delving into surrounding evidence.
Beyond Ethereum Research Circles
Buterin continues to test these ideas publicly, recently giving a demo Anonymous billboard Based on zero-knowledge proofs. The demo demonstrated how verifiable claims can move from research repositories to working products. Researchers have also begun formally validating agreed-upon Lean customers to detect errors early.
However, the trend reflects a familiar pattern: separate quick code from readable claims, and then prove that the two match.
There is no prototype for the new language yet, and Buterin has left the exact syntax open. Developers may agree on one common standard, or settle for several incompatible dialects instead. This choice can determine how quickly AI-verified code reaches production systems.
this post Vitalik Buterin proposes a language to make AI proofs readable appeared first on BeInCrypto.
