Reproduce “Beyond Prompt Injection: The Tool Update Gap in LLM Agent Security” 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
Shiyu Kan, Yue Qi, Yufei Han, Xinjian Luo, Guojun Xiong, Xiaofeng Gao
The authors study the tool update gap: approved tool code changes while recognized identity and authorization persist. In a controlled Markdown-cleaning test, an updated tool reads synthetic credentials and sends them to a local receiver. Across 21 framework–workload conditions, ETUDE achieves 100.0% attack success and 99.6% task completion, versus 20.2% and 96.4% for P-ctx×E2; the findings are limited to controlled tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.