Reproduce “Welcome to the Age of Agent Ads: Using On-Policy Attacks to Steer Agentic Decisions” 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
Tim R. Davidson, Marija Šakota, Joshua Joseph, Robert West
The authors test whether truthful absence claims steer agent rankings when paired with fabricated backstories about fictitious concepts. In ranking experiments, some models favored listings with the claims after receiving mocked search snippets; claims alone had minimal effect. In Beauty Products, the top-two rate rose from 0.39 to 0.56. This does not establish attacks on deployed agents or retrieval.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.