Reproduce “CyberPersistBench: Evaluating LLM-Based Cyber Attackers on Installation and Persistence” 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
Sujin Chen, Lijun Li, Xuhong Wang, Jing Shao
CyberPersistBench evaluates whether agents can establish host persistence from restricted post-compromise access. It tests artifacts through native registration, activation, credential revocation, and service, process, or host disruptions, using staged checks for fresh execution, native attribution, and host integrity. GLM-5.2’s cumulative rate falls from 58.6% at L1 to 43.3% at L2; the benchmark does not measure end-to-end attacks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.