The authors built nine agent configurations by pairing language models (open-weight and closed) with agentic frameworks (open-source and custom).
Open-weight agents pose a distinct and growing risk for LLM pollution in online surveys
Rilla et al. show that fully open, locally-run agent configurations match commercial systems in survey performance while evading different detection checks, making them harder to detect with existing methods
Research Lab
Raluca Rilla · Anne-Marie Nussberger · Rui Mata · Dirk U. Wulff
Max Planck Institute for Human Development · University of Basel · Vienna University of Economics and Business
Research Digest··2 min read
Thread:Agent Security & Attacks
The authors compared nine agent configurations, from fully open to closed commercial, on their ability to autonomously complete a survey and evade detection.
Why this paper
From Max Planck Institute for Human Development and 2 others · Part of Agent Security & Attacks, now 46 papers
In one line
Open-weight LLM agents running locally pose a distinct, low-cost threat for polluting human survey data.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§