ChainAgent searches PHP source for chained vulnerabilities and gadgets
Original title: ChainAgent: Automated Chainable Vulnerability Discovery in PHP Web Applications
On 30 PHP apps it reports 20 vulnerabilities and 13 exploit chains, nine reaching remote code execution.
ChainAgent is a goal-directed LLM-agent framework that jointly finds exploit-chain components and capability dependencies in PHP source, instead of analyzing flaws in isolation. An empirical study of 386 real-world web exploit chains guides searches over preconditions, capabilities, and functional gadgets. Against Codex on 15 PHP applications and three LLM backends, it reports 42.02%–44.63% recall for goal-reaching components and up to 4.4 times as many correct dependency edges.
- The comparison used 15 PHP applications and three shared LLM backends.
- Goal-reaching component recall was 42.02% to 44.63%.
- It recovered up to 4.4 times as many correct dependency edges as Codex.
- A further evaluation covered 30 real-world PHP applications.
From the original article
1 min read at arxiv.orgHigh-impact attacks on PHP web applications often chain limited-capability vulnerabilities with other flaws or intended behaviors. Discovering these chains requires identifying flaws and the dependencies enabling capability composition. Existing vulnerability discovery techniques analyze flaws in isolation, while attack planning assumes known components.
By Xinyi Wang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.