Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.29801 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916060879388672 |
|---|---|
| author | Liu, Dongrui Li, Yu Yang, Zhonghao Wang, Peng Chen, Guanxu Xie, Yuejin Mao, Qinghua Qu, Wanying Zhu, Yanxu Zhou, Tianyi Yuan, Leitao Zheng, Zhijie Lin, Qihao Wang, Yimin Luo, Haoyu Shao, Shuai Qian, Chen Liu, Qingyu Tang, Ling Qin, Ruiyang Ren, Qihan Yang, Junxiao Wang, Kun Xi, Zhiheng Zhang, Linfeng Duan, Ranjie Zhang, Bo Wang, Wenjie Shen, Wen Zhang, Qiaosheng Teng, Yan Lu, Chaochao Mei, Rui Li, Man Tao, Jialing Lin, Xi Zheng, Tianhang Liu, Yong Zhang, Quanshi Zhu, Lei Ma, Xingjun Liu, Junhua Xue, Hui Zuo, Xiaoxiang He, Xiangnan Shen, Chao Liu, Xianglong Huang, Minlie Shao, Jing Hu, Xia |
| author_facet | Liu, Dongrui Li, Yu Yang, Zhonghao Wang, Peng Chen, Guanxu Xie, Yuejin Mao, Qinghua Qu, Wanying Zhu, Yanxu Zhou, Tianyi Yuan, Leitao Zheng, Zhijie Lin, Qihao Wang, Yimin Luo, Haoyu Shao, Shuai Qian, Chen Liu, Qingyu Tang, Ling Qin, Ruiyang Ren, Qihan Yang, Junxiao Wang, Kun Xi, Zhiheng Zhang, Linfeng Duan, Ranjie Zhang, Bo Wang, Wenjie Shen, Wen Zhang, Qiaosheng Teng, Yan Lu, Chaochao Mei, Rui Li, Man Tao, Jialing Lin, Xi Zheng, Tianhang Liu, Yong Zhang, Quanshi Zhu, Lei Ma, Xingjun Liu, Junhua Xue, Hui Zuo, Xiaoxiang He, Xiangnan Shen, Chao Liu, Xianglong Huang, Minlie Shao, Jing Hu, Xia |
| contents | Modern open-world agents such as OpenClaw exhibit powerful cross-environment execution capabilities yet introduce broad new safety risk sources. Meanwhile, advanced frontier AI models drastically lower attack barriers, rendering current agent alignment frameworks inadequate for real-world deployment. To tackle these emerging threats, we propose a lightweight and scalable agent safety alignment framework. Specifically, we update the agent safety taxonomy to accommodate emergent risks from Codex and OpenClaw execution scenarios. We further build a taxonomy-guided data engine with influence-function purification to train lightweight AgentDoG 1.5 variants (0.8B, 2B, 4B, and 8B parameters) using only around 1k samples, achieving comparable performance with leading closed-source models (e.g., GPT-5.4). Based on AgentDoG 1.5, we construct a highly efficient agentic safety SFT and RL training environment, which reduces deployment overhead in Docker-level environments by two orders of magnitude. Finally, we deploy AgentDoG 1.5 as a training-free online guardrail for real-time safety moderation. Extensive experimental results indicate that AgentDoG 1.5 achieves state-of-the-art performance in diverse and complex interactive agentic scenarios. All models and datasets are openly released. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_29801 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security Liu, Dongrui Li, Yu Yang, Zhonghao Wang, Peng Chen, Guanxu Xie, Yuejin Mao, Qinghua Qu, Wanying Zhu, Yanxu Zhou, Tianyi Yuan, Leitao Zheng, Zhijie Lin, Qihao Wang, Yimin Luo, Haoyu Shao, Shuai Qian, Chen Liu, Qingyu Tang, Ling Qin, Ruiyang Ren, Qihan Yang, Junxiao Wang, Kun Xi, Zhiheng Zhang, Linfeng Duan, Ranjie Zhang, Bo Wang, Wenjie Shen, Wen Zhang, Qiaosheng Teng, Yan Lu, Chaochao Mei, Rui Li, Man Tao, Jialing Lin, Xi Zheng, Tianhang Liu, Yong Zhang, Quanshi Zhu, Lei Ma, Xingjun Liu, Junhua Xue, Hui Zuo, Xiaoxiang He, Xiangnan Shen, Chao Liu, Xianglong Huang, Minlie Shao, Jing Hu, Xia Artificial Intelligence Computation and Language Cryptography and Security Computer Vision and Pattern Recognition Machine Learning Modern open-world agents such as OpenClaw exhibit powerful cross-environment execution capabilities yet introduce broad new safety risk sources. Meanwhile, advanced frontier AI models drastically lower attack barriers, rendering current agent alignment frameworks inadequate for real-world deployment. To tackle these emerging threats, we propose a lightweight and scalable agent safety alignment framework. Specifically, we update the agent safety taxonomy to accommodate emergent risks from Codex and OpenClaw execution scenarios. We further build a taxonomy-guided data engine with influence-function purification to train lightweight AgentDoG 1.5 variants (0.8B, 2B, 4B, and 8B parameters) using only around 1k samples, achieving comparable performance with leading closed-source models (e.g., GPT-5.4). Based on AgentDoG 1.5, we construct a highly efficient agentic safety SFT and RL training environment, which reduces deployment overhead in Docker-level environments by two orders of magnitude. Finally, we deploy AgentDoG 1.5 as a training-free online guardrail for real-time safety moderation. Extensive experimental results indicate that AgentDoG 1.5 achieves state-of-the-art performance in diverse and complex interactive agentic scenarios. All models and datasets are openly released. |
| title | AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security |
| topic | Artificial Intelligence Computation and Language Cryptography and Security Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2605.29801 |