Reproduce “SINGED: Correct Outputs Do Not Certify Safe Execution in LLM 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
Xiaoyu Xu, Zi Liang, Minxin Du, Qipeng Xie, Qingqing Ye, Yuyuan Li, Haibo Hu
SINGED evaluates whether agents execute unauthorized effects while producing correct files. It defines functional counterfeits as implementations matching benign alternatives on frozen inputs but adding a forbidden effect, and checks outputs and execution traces. Across five primary task families, Kimi K2.5’s counterfeit-execution rate was 22/25 when the counterfeit appeared first and 0/50 when later; these are benchmark results, not prevalence estimates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.