The non-profit Ethereum Foundation has launched better codes – an open research competition in which AI agents will compete to improve the mathematically proven security of the koalaIRS12 cryptographic problem. The initiative should bring the proven lower bound of security closer to 128 bits and strengthen the foundation for SNARK systems, which are used in zk-rollups, zkVM, and Ethereum’s post‑quantum infrastructure.

The project was created by the formal verification team of the Ethereum Foundation in collaboration with Yukon and zkSecurity. Participants can connect their own AI models, tools, and agent management systems to the common task, and each result is verified by the Lean core. The basis of the competition was the koalaIRS12 task, which originates from Proximity Prize research and is related to Reed–Solomon codes. It was fully formalized in Lean, and the proven security metric was made public in the rating.

Participants must increase the bounds of the problem’s correctness, i.e., the proven lower bound of its cryptographic security. Each submission is verified by the Lean kernel, and the accepted results are added to the public repository. Each improved proof contains new lemmas, proof methods, and impossibility results, which are then added to the main repository to accelerate progress for all solvers and agents.

Unlike the closed study of 1 team, the better codes model предполагает parallel work of a large number of independent AI agents on 1 verified benchmark. Participants can log in via GitHub, clone the repository, and submit their own proofs. The accepted results are published with the author and the AI model used indicated. This allows other participants to use new lemmas and methods and avoid repeating already unsuccessful approaches.

The context for the launch is Ethereum’s broader work on post‑quantum cryptography. After 8 years of research, the Ethereum Foundation decided to abandon the specialized Poseidon hash in favor of SHA and BLAKE, as new SNARK constructions have shown that traditional hash functions can provide comparable efficiency. The better codes contest will be another step in the development of provably secure cryptographic systems, which will form the foundation of the next generation of Ethereum.