Papers & Research

ExploitGym: Can AI agents turn security vulnerabilities into real attacks?

Original title:ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks?

AI summary

A team of researchers from Berkeley RDI, the Max Planck Institute for Security and Privacy, Anthropic, OpenAI, Google and other institutions released a new benchmark called ExploitGym. The benchmark contains 898 real vulnerabilities and requires AI agents to generate complete exploits based on vulnerability descriptions. Results show that frontier AI models can already successfully exploit a considerable number of vulnerabilities, and some attacks still succeed even with standard defenses such as ASLR enabled. This demonstrates that AI already has the ability to autonomously turn vulnerabilities into real attacks, a dual-use capability that can help defenders assess vulnerability severity while also potentially lowering the technical barrier for attackers.

Key Takeaways

  • A team of researchers from Berkeley RDI,the Max Planck Institute for Security and…
  • The benchmark contains 898 real vulnerabilities and requires AI agents to generate complete…
  • Results show that frontier AI models can already successfully exploit a considerable number of…
  • This demonstrates that AI already has the ability to autonomously turn vulnerabilities into real…

Quick Facts

Published
Updated
Source
Berkeley RDI:Blog(AI 安全与评测)
Category
Papers & Research
AI score
79
Read the original Read the full article at the original source
Why it matters

顶级 AI 模型已能自己把已知软件漏洞变成可运行攻击代码,连 ASLR 等标准防御都挡不住部分攻击,研究更发现模型会主动寻找更危险的意外漏洞。安全行业不能再把这当成假设性问题了。

Related entities

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.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: ExploitGym: Can AI agents turn security vulnerabilities into real attacks?. May 13, 2026. https://new.nat6.net/en/news/cmp4j0582068osljxjmrn48ll/

View all →