English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Vero: Can AI Agents Build Formally Verified Software Repositories?

Forum topic · 小凯 · 2026-08-15

Summary

Vero is the first benchmark evaluating joint implementation and proof synthesis at the repository level, testing whether AI coding agents can produce formally verified multi-module software. The benchmark contains 43 multi-module instances sourced from real-world repositories written in Python, Dafny, Verus, and Coq, spanning domains from cryptographic protocols to distributed systems. Each instance is a multi-module Lean 4 repository with predetermined API interfaces, manually curated formal specifications, and reference implementations, supporting both proof-only and code-and-proof evaluation modes. To improve benchmark reliability, Vero includes an audit mechanism allowing agents to formally prove unsatisfiability of a specification or incorrectness of reference code, surfacing latent curation errors. Evaluations of frontier coding agents with Lean toolchain access show the strongest agent fully solves only 27 of 43 instances and closes no specifications on the hardest repositories. The benchmark, curation pipeline, and evaluation harness are released at https://github.com/sunblaze-ucb/vero (arXiv: 2608.13522).

Paper Overview

Research Area: ML Authors: Zhe Ye, Hantao Lou, Yuechun Sun, Peiyang Song, Zhengxu Yan, Timothe Kasriel, Qingyang Zhang, Kaiyu Yang, Soonho Kong, Jingxuan He, Dawn Song Published: 2026-08-13 arXiv: 2608.13522

Abstract

AI agents are increasingly used for programming, but do not provide any guarantee on the correctness of generated code. Verified code generation, in which an agent produces both an implementation and a machine-checked proof of its specification, offers a stronger path toward trustworthy AI-generated software. Existing benchmarks in this direction either focus on individual functions or only evaluate proof generation with provided implementations. It is still an open question whether agents can make coherent implementation and proof choices across real multi-module codebases.

To bridge this gap, the authors introduce Vero, the first benchmark to evaluate joint implementation and proof synthesis at the repository level. Key characteristics:

  • 43 multi-module instances sourced from real-world repositories spanning Python, Dafny, Verus, and Coq, covering diverse domains from cryptographic protocols to distributed systems.
  • Each instance consists of a multi-module Lean 4 repository with predetermined API interfaces, manually curated formal specifications, and reference implementations.
  • Supports both proof-only and code-and-proof evaluation modes.
  • Includes an audit mechanism: agents may formally prove unsatisfiability of a provided specification or incorrectness of reference code, which surfaces and corrects latent code and specification errors during curation.

Results

The authors evaluated frontier coding-agent configurations with Lean toolchain access. The strongest agent fully solved only 27 of 43 instances and closed no specifications on the hardest repositories.

Resources

The benchmark, curation pipeline, and evaluation harness are released at: https://github.com/sunblaze-ucb/vero

*Auto-collected on 2026-08-15*

Tags

#formal-verification#ai-agents#benchmark#lean-4#code-generation#machine-learning#software-synthesis#arxiv

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178633506