Reproduce “CyberClear: A Benchmark for LLM Agent Systems on APT Attack Chain Provenance” 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
Qi Chen, Fushuo Huo, Hangli Shen, Jingcai Guo, Shuhao Li, Guang Cheng
CyberProvenance reconstructs attack-chain provenance from security logs by preserving accumulated evidence, testing proposed techniques in isolated environments, and revising unsupported links. CyberClear evaluates 450 instances using five semantic dimensions. With GPT-5.6-Terra, CyberProvenance scores 0.8897 Single, 0.7438 Loose, 0.6337 Strict, 0.6124 Detail, and 0.7263 Scenario; DyLAN leads Loose. These results are for this benchmark, and validation does not prove historical causation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.