SafeHarbor: Hierarchical Memory-Augmented Guardrail for LLM Agent Safety
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914589710483456 |
|---|---|
| author | Liu, Zhe Ying, Zonghao Zhang, Wenxin Zou, Quanchen Zhang, Deyue Yang, Dongdong Zhang, Xiangzheng Peng, Hao |
| author_facet | Liu, Zhe Ying, Zonghao Zhang, Wenxin Zou, Quanchen Zhang, Deyue Yang, Dongdong Zhang, Xiangzheng Peng, Hao |
| contents | Recent advances in foundation models have transformed LLMs from passive conversational systems into autonomous agents capable of reasoning and tool execution. While these capabilities unlock substantial practical value, they also introduce new security risks, as adversaries can manipulate agents into performing harmful actions in real-world environments. Existing defense strategies mitigate such threats but frequently struggle to balance safety and utility, resulting in over-refusal of benign user requests. 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_05704 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SafeHarbor: Hierarchical Memory-Augmented Guardrail for LLM Agent Safety Liu, Zhe Ying, Zonghao Zhang, Wenxin Zou, Quanchen Zhang, Deyue Yang, Dongdong Zhang, Xiangzheng Peng, Hao Cryptography and Security Artificial Intelligence 68T42 Recent advances in foundation models have transformed LLMs from passive conversational systems into autonomous agents capable of reasoning and tool execution. While these capabilities unlock substantial practical value, they also introduce new security risks, as adversaries can manipulate agents into performing harmful actions in real-world environments. Existing defense strategies mitigate such threats but frequently struggle to balance safety and utility, resulting in over-refusal of benign user requests. 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. |
| title | SafeHarbor: Hierarchical Memory-Augmented Guardrail for LLM Agent Safety |
| topic | Cryptography and Security Artificial Intelligence 68T42 |
| url | https://arxiv.org/abs/2605.05704 |