BadAgent Extension

Abstract

This paper presents an empirical study on backdoor attacks in large language model agents. We extend a recent attack framework by adding two lightweight benchmarks that measure cross-domain robustness and trigger visibility without changing the model architecture. Our approach fine-tunes AgentLM-based agents with parameter-efficient methods on operating system and web browsing tasks using multiple poisoning ratios and both visible and invisible triggers. We then evaluate the agents with

Categories

Framework mappings

Suggested from the entry's categories.

@misc{garrido2026badagent,
  title = {{BadAgent Extension}},
  author = {Pedro Yanes Garrido and Diego Fernandez Arias},
  year = {2026},
  month = jun,
  doi = {10.4018/979-8-3373-8252-4.ch010},
  url = {https://doi.org/10.4018/979-8-3373-8252-4.ch010}
}