Reproduce “RapidPen: Fully Automated IP-to-Shell Penetration Testing with LLM-based Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sho Nakatani
RapidPen, developed by Sho Nakatani at SecDevLab Inc., introduces a fully automated framework for IP-to-Shell penetration testing, leveraging LLM-based agents to achieve initial system access without human intervention. The system demonstrated a 60% success rate on a vulnerable target, completing tasks within 200-400 seconds at an average LLM cost of $0.3-$0.6 per run, particularly when guided by pre-recorded 'success cases'.
Key insights
- Reusing 'success cases' from past exploits significantly improves the success rate and consistency of automated initial access (60% vs. 30% without).
- Iterative command refinement, including retries and dynamic tool replacement (e.g., `nmap` to `rustscan`), enhances the system's ability to overcome execution failures and achieve exploitation.
- The ReAct paradigm, augmented by domain-specific RAGs, enables LLMs to effectively plan and execute complex, multi-stage offensive security tasks autonomously.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.