Reproduce “PROVISO: Provenance-Informed Action Selection for Secure 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
Bo Yan, Qi Zhang, Ziwen Kan, Fnu Suya, Song Wang
PROVISO selects actions for agents by tracking information provenance and comparing a proposed action with an available alternative. Its learned controller uses trusted context to constrain objectives while external information can support action arguments. On ToolSafe, unsafe-call F1—the harmonic mean of precision and recall—was 99.12%, versus 92.05% for TS-Guard. Controllers are fitted separately per benchmark; this result does not guarantee deployment safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.