titleSafeHarbor: Hierarchical Memory-Augmented Guardrail for LLM Agent Safety
authorsZhe Liu, Zonghao Ying, Wenxin Zhang, Quanchen Zou, Deyue Zhang, Dongdong Yang, Xiangzheng Zhang, Hao Peng
date2026-05-07
venuearXiv (preprint, not peer reviewed)
idarXiv:2605.05704
affiliationBeihang University
upvotes1 (Hugging Face Daily Papers, 2026-11-01)
linksabs pdf hf code
primerMemory for AI agents
abstractWith the rapid evolution of foundation models, Large Language Model (LLM) agents have demonstrated increasingly powerful tool-use capabilities. However, this proficiency introduces significant security risks, as malicious actors can manipulate agents into executing tools to generate harmful content. While existing defensive mechanisms are effective, they frequently suffer from the over-refusal problem, where increased safety strictness compromises the agent's utility on benign tasks. To mitigate this trade-off, we propose SafeHarbor, a novel framework designed to establish precise decision boundaries for LLM agents. Unlike static guidelines, SafeHarbor extracts context-aware defense rules through enhanced adversarial generation. We design a local hierarchical memory system for dynamic rule injection, offering a training-free, efficient, and plug-and-play solution. Furthermore, we introduce an information entropy-based self-evolution mechanism that continuously optimizes the memory structure through dynamic node splitting and merging. Extensive experiments demonstrate that SafeHarbor achieves state-of-the-art performance on both ambiguous benign tasks and explicit malicious attacks, notably attaining a peak benign utility of 63.6\% on GPT-4o while maintaining a robust refusal rate exceeding 93\% against harmful requests. The source code is publicly available at https://github.com/ljj-cyber/SafeHarbor.

cite

@misc{liu2026safeharbor,
  title={SafeHarbor: Hierarchical Memory-Augmented Guardrail for LLM Agent Safety},
  author={Zhe Liu and Zonghao Ying and Wenxin Zhang and Quanchen Zou and Deyue Zhang and Dongdong Yang and Xiangzheng Zhang and Hao Peng},
  year={2026},
  eprint={2605.05704},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2605.05704}
}