The Ethereum Foundation Formal Verification team, working in close collaboration with Yukon and zkSecurity, has officially announced the launch of better.codes, an open autoresearch platform designed to tackle complex mathematical problems in cryptography. The new initiative invites researchers and developers to deploy autonomous artificial intelligence agents to push the machine-checked soundness bounds of cryptographic primitives used in modern zero-knowledge proof systems. By leveraging formal verification tools and open public leaderboards, the project aims to bridge the historical gap between conjectured security benchmarks and mathematically proven guarantees.
At the core of the initial launch is a self-contained research problem derived from the Proximity Prize initiative, which was established earlier this year to address fundamental questions in list decoding and correlated agreement. Specifically, the better.codes platform centers on koalaIRS12, a Reed–Solomon proximity problem designed to advance modern succinct non-interactive argument of knowledge systems, commonly known as SNARKs. Solvers participating in the challenge direct their custom AI agents to raise the machine-checked soundness bound of this specific problem, working incrementally toward a fixed 128-bit target that is standard in production cryptographic systems.
Why Provable Bits Matter in Cryptography
To understand the significance of the better.codes challenge, one must examine the foundational architecture of modern cryptographic protocols. Nearly all production hash-based SNARKs—ranging from the proof systems that secure decentralized zkrollups and zero-knowledge virtual machines (zkVMs) to the advanced architectures central to Ethereum’s long-term post-quantum roadmap—rely heavily on proximity gaps and correlated agreement for Reed–Solomon codes. These mathematical structures allow verifiers to be convinced that a prover possesses a valid witness without revealing the underlying data.
However, a persistent challenge in applied cryptography is that what can be rigorously proven about these mathematical systems today often stops short of what researchers believe the benchmarks actually are. Deployed production systems typically target a robust 128-bit security level. Yet, that vital security guarantee holds in full only if the underlying mathematical conjectures are entirely correct.
The better.codes autoresearch challenge is specifically engineered to close this gap between conjectured security benchmarks and proven security benchmarks. By fostering an open, incremental, verifiable, and public research environment, the initiative encourages the mathematical community to turn theoretical assumptions into undeniable proofs. The core problem, koalaIRS12, bridges directly to the grand challenges laid out earlier this year in the academic paper "Open Problems in List Decoding and Correlated Agreement," authored by Gal Arnon, Dan Boneh, and Giacomo Fenzi. Furthermore, the problem is fully formalized end to end within ArkLib, the specialized Lean 4 library for formally verified arguments of knowledge.
An Always-On Autoresearch Model
The launch of better.codes introduces a novel collaborative paradigm known as "always-on autoresearch." Rather than relying on traditional, isolated academic research cycles or single centralized teams, this model enables participants from across the global cryptographic and artificial intelligence communities to run their own AI models, experimental harnesses, and automated tools in parallel against a common, highly rigorous verified benchmark.
Because no single agentic setup or automated architecture is universally optimal across an open and complex mathematical problem, deploying many independent setups working simultaneously on the same benchmark moves the research frontier forward at an unprecedented pace. This decentralized, swarm-like approach has already proven successful in previous open research initiatives. Similar collaborative platforms, including ecdsa.fail, zk.golf, and snark.fast, have successfully advanced research frontiers in fields such as quantum circuit design, verified zero-knowledge circuits, and post-quantum proving speeds.
In this ecosystem, progress is cumulative. When an automated agent or human researcher discovers a breakthrough, the Lean kernel rigorously checks every aspect of the submission. Each promoted proof successfully raises the floor for all subsequent participants, ensuring that the entire community benefits from the collective intelligence brought to the platform.
How the Verification and Submission Process Works
Participating in the better.codes challenge is designed to be accessible to researchers equipped with modern development tools while maintaining rigorous standards of cryptographic verification. Solvers begin by signing in with GitHub directly at the better.codes platform and cloning the official challenge repository. Within the repository, the overarching theorem statement, the specific parameter point, and the verification harness are permanently pinned to ensure consistency.
Solvers then work within a designated submission surface, utilizing their AI agents and local infrastructure to prove a larger soundness lower bound, which is scored precisely in bits. Once a potential proof is developed, a specialized comparator checks that the submission’s exported theorem matches the pinned statement exactly. Following this structural check, the Lean kernel independently verifies the mathematical validity of the proof.
Accepted results are promptly promoted to the public repository, where full credit is publicly attributed to both the human solver and the specific AI model or harness used to generate the solution. Transparency is a foundational pillar of the platform; all submissions are entirely transparent and git-backed. As new lemmas, innovative proof techniques, and critical impossibility results are discovered, they are upstreamed directly into the main repository. This allows any subsequent solver or AI agent to review past diffs and submission notes, build directly upon prior work, and systematically avoid dead ends, thereby accelerating progress for the entire ecosystem.
Looking Ahead at the Future of Verified Cryptography
Today’s official launch covers the foundational soundness challenge aimed at raising the proven lower bound for the koalaIRS12 problem up to the coveted 128-bit mark. As the initiative gains momentum, the organizers hope to introduce additional cryptographic challenges over time, expanding the scope of automated formal verification into other critical areas of zero-knowledge research.
Participation, evaluation criteria, awards, and any associated disbursements are governed by the established program terms of the challenge, which may be adjusted as the research initiative progresses and evolves. Researchers, cryptographers, and AI developers interested in contributing to the future of provable security can access the repository and begin working on the challenge directly through the official better.codes portal.
Comments (0)