Berkeley RDI releases CyberBinGym benchmark to test whether AI agents can write real exploits from binary patches alone
Original title:CyberBinGym: Can AI Agents Turn Binary Patches into Real Exploits?
AI summary
UC Berkeley and other institutions released the CyberBinGym benchmark to evaluate whether AI agents can recover and exploit real vulnerabilities in 1-day scenarios with no source code, no vulnerability description, and no trigger input, using only stripped vulnerable and patched binaries. It includes 100 tasks from 53 open-source projects.
Key Takeaways
- UC Berkeley and other institutions released the CyberBinGym benchmark to evaluate whether AI…
- no vulnerability description,and no trigger input,using only stripped vulnerable and patched…
- It includes 100 tasks from 53 open-source projects.
Quick Facts
- Published
- Source
- Berkeley RDI:Blog(AI 安全与评测)
- Category
- Papers & Research
- Topic
- AgentSafety & Alignment
- AI score
- 61
基准把漏洞恢复与可审计利用分开度量,结果显示 Codex 恢复 88% 但仅 20% 能完成利用,硬化编译进一步压缩成功率。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Berkeley RDI releases CyberBinGym benchmark to test whether AI agents can write real exploits from binary patches alone. November 11, 2026. https://new.nat6.net/en/news/xenhfme3azzcf8qpihr57fb93/