_version_ 1866908988667330560
author Liu, Dongrui
Ren, Qihan
Qian, Chen
Shao, Shuai
Xie, Yuejin
Li, Yu
Yang, Zhonghao
Luo, Haoyu
Wang, Peng
Liu, Qingyu
Hu, Binxin
Tang, Ling
Mei, Jilin
Guo, Dadi
Yuan, Leitao
Yang, Junyao
Chen, Guanxu
Lin, Qihao
Yu, Yi
Zhang, Bo
Guo, Jiaxuan
Zhang, Jie
Shao, Wenqi
Deng, Huiqi
Xi, Zhiheng
Wang, Wenjie
Wang, Wenxuan
Shen, Wen
Chen, Zhikai
Xie, Haoyu
Tao, Jialing
Dai, Juntao
Ji, Jiaming
Ba, Zhongjie
Zhang, Linfeng
Liu, Yong
Zhang, Quanshi
Zhu, Lei
Wei, Zhihua
Xue, Hui
Lu, Chaochao
Shao, Jing
Hu, Xia
author_facet Liu, Dongrui
Ren, Qihan
Qian, Chen
Shao, Shuai
Xie, Yuejin
Li, Yu
Yang, Zhonghao
Luo, Haoyu
Wang, Peng
Liu, Qingyu
Hu, Binxin
Tang, Ling
Mei, Jilin
Guo, Dadi
Yuan, Leitao
Yang, Junyao
Chen, Guanxu
Lin, Qihao
Yu, Yi
Zhang, Bo
Guo, Jiaxuan
Zhang, Jie
Shao, Wenqi
Deng, Huiqi
Xi, Zhiheng
Wang, Wenjie
Wang, Wenxuan
Shen, Wen
Chen, Zhikai
Xie, Haoyu
Tao, Jialing
Dai, Juntao
Ji, Jiaming
Ba, Zhongjie
Zhang, Linfeng
Liu, Yong
Zhang, Quanshi
Zhu, Lei
Wei, Zhihua
Xue, Hui
Lu, Chaochao
Shao, Jing
Hu, Xia
contents The rise of AI agents introduces complex safety and security challenges arising from autonomous tool use and environmental interactions. Current guardrail models lack agentic risk awareness and transparency in risk diagnosis. To introduce an agentic guardrail that covers complex and numerous risky behaviors, we first propose a unified three-dimensional taxonomy that orthogonally categorizes agentic risks by their source (where), failure mode (how), and consequence (what). Guided by this structured and hierarchical taxonomy, we introduce a new fine-grained agentic safety benchmark (ATBench) and a Diagnostic Guardrail framework for agent safety and security (AgentDoG). AgentDoG provides fine-grained and contextual monitoring across agent trajectories. More Crucially, AgentDoG can diagnose the root causes of unsafe actions and seemingly safe but unreasonable actions, offering provenance and transparency beyond binary labels to facilitate effective agent alignment. AgentDoG variants are available in three sizes (4B, 7B, and 8B parameters) across Qwen and Llama model families. Extensive experimental results demonstrate that AgentDoG achieves state-of-the-art performance in agentic safety moderation in diverse and complex interactive scenarios. All models and datasets are openly released.
format Preprint
id arxiv_https___arxiv_org_abs_2601_18491
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security
Liu, Dongrui
Ren, Qihan
Qian, Chen
Shao, Shuai
Xie, Yuejin
Li, Yu
Yang, Zhonghao
Luo, Haoyu
Wang, Peng
Liu, Qingyu
Hu, Binxin
Tang, Ling
Mei, Jilin
Guo, Dadi
Yuan, Leitao
Yang, Junyao
Chen, Guanxu
Lin, Qihao
Yu, Yi
Zhang, Bo
Guo, Jiaxuan
Zhang, Jie
Shao, Wenqi
Deng, Huiqi
Xi, Zhiheng
Wang, Wenjie
Wang, Wenxuan
Shen, Wen
Chen, Zhikai
Xie, Haoyu
Tao, Jialing
Dai, Juntao
Ji, Jiaming
Ba, Zhongjie
Zhang, Linfeng
Liu, Yong
Zhang, Quanshi
Zhu, Lei
Wei, Zhihua
Xue, Hui
Lu, Chaochao
Shao, Jing
Hu, Xia
Artificial Intelligence
Computational Complexity
Computation and Language
Computer Vision and Pattern Recognition
Machine Learning
The rise of AI agents introduces complex safety and security challenges arising from autonomous tool use and environmental interactions. Current guardrail models lack agentic risk awareness and transparency in risk diagnosis. To introduce an agentic guardrail that covers complex and numerous risky behaviors, we first propose a unified three-dimensional taxonomy that orthogonally categorizes agentic risks by their source (where), failure mode (how), and consequence (what). Guided by this structured and hierarchical taxonomy, we introduce a new fine-grained agentic safety benchmark (ATBench) and a Diagnostic Guardrail framework for agent safety and security (AgentDoG). AgentDoG provides fine-grained and contextual monitoring across agent trajectories. More Crucially, AgentDoG can diagnose the root causes of unsafe actions and seemingly safe but unreasonable actions, offering provenance and transparency beyond binary labels to facilitate effective agent alignment. AgentDoG variants are available in three sizes (4B, 7B, and 8B parameters) across Qwen and Llama model families. Extensive experimental results demonstrate that AgentDoG achieves state-of-the-art performance in agentic safety moderation in diverse and complex interactive scenarios. All models and datasets are openly released.
title AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security
topic Artificial Intelligence
Computational Complexity
Computation and Language
Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2601.18491