PoCForge: A closed-loop multi-agent framework for autonomous exploit synthesis and verification

doi:10.1016/j.asoc.2026.116684

PoCForge: A closed-loop multi-agent framework for autonomous exploit synthesis and verification

2026-10-17

0

PRE

AI

DOI：10.1016/j.asoc.2026.116684

Original

Original request for help

Save

Abstract

En 中文

• PoCForge: multi-agent framework for closed-loop exploit synthesis and verification. • Buffer-of-Thoughts and three-tier hierarchy for robust PoC generation. • Achieves SOTA results: 52.5% on CVE-Bench and 100% on AutoPenBench. • Performs well with open-source LLMs via verified exploitation feedback. • Enables adaptive exploit synthesis beyond public PoCs across environments.

Keywords:

Penetration testing

Multi-agent system

Large language model

Exploit planning

Proof-of-concept generation

PoCForge: A closed-loop multi-agent framework for autonomous exploit synthesis and verification

Abstract

Journal

Organization

Cited Papers