**Title: SoK: DARPA's AI Cyber Challenge (AIxCC): Competition Design, Architectures, and Lessons Learned

Abstract: Running from 2023 to 2025, DARPA’s AI Cyber Challenge (AIxCC) stands as the most extensive competition to date focused on developing fully autonomous cyber reasoning systems (CRSs). These systems harness recent AI breakthroughs, specifically large language models (LLMs), to identify and patch vulnerabilities within real-world open-source software. This study offers the first comprehensive evaluation of the AIxCC initiative. By synthesizing design documentation, source code, execution logs, and insights gathered from both organizers and participating teams, we scrutinize the competition’s framework and pivotal design choices. Furthermore, we categorize the architectural strategies employed by the finalist CRSs and delve into performance outcomes that extend beyond the final rankings. Our findings highlight the critical determinants of CRS efficacy, document substantive technical progress made by the teams, and uncover persistent limitations that warrant further investigation. The paper concludes with recommendations for structuring future contests and provides broader perspectives on the practical deployment of autonomous CRSs.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC