ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection

Abstract

The rise of Large Language Model (LLM) agents, augmented with tool use, skills, and external knowledge, has introduced new security risks. Among them, prompt injection attacks, where adversaries embed malicious instructions into the agent workflow, have emerged as the primary threat. However, existing benchmarks and defenses are fundamentally limited as they assume context-insensitive settings in which the agent works under a fully specified user instruction, and the attacks are straightforward

Categories

Framework mappings

Suggested from the entry's categories.

@misc{weng2026argus,
  title = {{ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection}},
  author = {Shihao Weng and Yang Feng and Jinrui Zhang and Xiaofei Xie and Jiongchi Yu and Jia Liu},
  year = {2026},
  month = may,
  eprint = {2605.03378},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2605.03378}
}