Reproduce “Securing Web Agents with the Compile-Then-Execute Paradigm” 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
Annabella Chow, Jinhao Zhu, Sylvie Venuto, Abdulaziz Alharbi, Basel Alomair, Raluca Popa, David Wagner, Julien Piet
Baya compiles a web task into a program before the agent accesses website content, then executes it through a trusted website API. This keeps untrusted content from adding operations, while allowing it to be processed as data. Across 581 WebArena Verified tasks, Baya utility reached 77.8%, versus 66.7% for API-enabled ReAct; tested attacks had 0% observed ASR, not a universal guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.