Reproduce “APEX: Active Protection at Execution Boundaries for 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
Xinran Zheng, Xin Fan Guo, Zhiqiang Hao, Fan Yang, Xingzhi Qian, Jiawei Du, Jinfeng Xu, Zheng Xing and 2 more
APEX checks agent actions against a contract before external effects are committed. The contract records permitted effects and endorsed sources for action arguments; WRAP reconstructs proposals from permitted evidence, while PLANT probes for unendorsed influence. On ASB-OPI, APEX reports 0% attack success, versus 9.80% for DRIFT, with 90.20% benign utility. Its protection depends on faithful contracts and registered interfaces.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.