A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security</p>\n","updatedAt":"2026-05-29T02:51:13.754Z","author":{"_id":"6621f4eb64e84619e578aad6","avatarUrl":"/avatars/b1ad96ee354b999fcafb2998a636609c.svg","fullname":"Dongrui Liu","name":"shenqiorient","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8979578614234924},"editors":["shenqiorient"],"editorAvatarUrls":["/avatars/b1ad96ee354b999fcafb2998a636609c.svg"],"reactions":[],"isReport":false}},{"id":"6a1a4080ca63123d87cefd64","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":359,"isUserFollowing":false},"createdAt":"2026-05-30T01:42:24.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [Orchard: An Open-Source Agentic Modeling Framework](https://huggingface.co/papers/2605.15040) (2026)\n* [Auditing Agent Harness Safety](https://huggingface.co/papers/2605.14271) (2026)\n* [SafeHarbor: Hierarchical Memory-Augmented Guardrail for LLM Agent Safety](https://huggingface.co/papers/2605.05704) (2026)\n* [Security Risks in Tool-Enabled AI Agents: A Systematic Analysis of Privileged Execution Environments](https://huggingface.co/papers/2605.09721) (2026)\n* [A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework](https://huggingface.co/papers/2604.23338) (2026)\n* [ADR: An Agentic Detection System for Enterprise Agentic AI Security](https://huggingface.co/papers/2605.17380) (2026)\n* [The Blind Spot of Agent Safety: How Benign User Instructions Expose Critical Vulnerabilities in Computer-Use Agents](https://huggingface.co/papers/2604.10577) (2026)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"<p>This is an automated message from the <a href=\"https://huggingface.co/librarian-bots\">Librarian Bot</a>. I found the following papers similar to this paper. </p>\n<p>The following papers were recommended by the Semantic Scholar API </p>\n<ul>\n<li><a href=\"https://huggingface.co/papers/2605.15040\">Orchard: An Open-Source Agentic Modeling Framework</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.14271\">Auditing Agent Harness Safety</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.05704\">SafeHarbor: Hierarchical Memory-Augmented Guardrail for LLM Agent Safety</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.09721\">Security Risks in Tool-Enabled AI Agents: A Systematic Analysis of Privileged Execution Environments</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.23338\">A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2605.17380\">ADR: An Agentic Detection System for Enterprise Agentic AI Security</a> (2026)</li>\n<li><a href=\"https://huggingface.co/papers/2604.10577\">The Blind Spot of Agent Safety: How Benign User Instructions Expose Critical Vulnerabilities in Computer-Use Agents</a> (2026)</li>\n</ul>\n<p> Please give a thumbs up to this comment if you found it helpful!</p>\n<p> If you want recommendations for any Paper on Hugging Face checkout <a href=\"https://huggingface.co/spaces/librarian-bots/recommend_similar_papers\">this</a> Space</p>\n<p> You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: <code><span class=\"SVELTE_PARTIAL_HYDRATER contents\" data-target=\"UserMention\" data-props=\"{&quot;user&quot;:&quot;librarian-bot&quot;}\"><span class=\"inline-block\"><span class=\"contents\"><a href=\"/librarian-bot\">@<span class=\"underline\">librarian-bot</span></a></span> </span></span> recommend</code></p>\n","updatedAt":"2026-05-30T01:42:24.138Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":359,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7084611654281616},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.29801","authors":[{"_id":"6a18fe4056b4bb14ec65cefc","name":"Dongrui Liu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cefd","name":"Yu Li","hidden":false},{"_id":"6a18fe4056b4bb14ec65cefe","name":"Zhonghao Yang","hidden":false},{"_id":"6a18fe4056b4bb14ec65ceff","name":"Peng Wang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf00","name":"Guanxu Chen","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf01","name":"Yuejin Xie","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf02","name":"Qinghua Mao","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf03","name":"Wanying Qu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf04","name":"Yanxu Zhu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf05","name":"Tianyi Zhou","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf06","name":"Leitao Yuan","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf07","name":"Zhijie Zheng","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf08","name":"Qihao Lin","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf09","name":"Yimin Wang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf0a","name":"Haoyu Luo","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf0b","name":"Shuai Shao","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf0c","name":"Chen Qian","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf0d","name":"Qingyu Liu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf0e","name":"Ling Tang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf0f","name":"Ruiyang Qin","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf10","name":"Qihan Ren","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf11","name":"Junxiao Yang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf12","name":"Kun Wang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf13","name":"Zhiheng Xi","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf14","name":"Linfeng Zhang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf15","name":"Ranjie Duan","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf16","name":"Bo Zhang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf17","name":"Wenjie Wang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf18","name":"Wen Shen","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf19","name":"Qiaosheng Zhang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf1a","name":"Yan Teng","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf1b","name":"Chaochao Lu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf1c","name":"Rui Mei","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf1d","name":"Man Li","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf1e","name":"Jialing Tao","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf1f","name":"Xi Lin","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf20","name":"Tianhang Zheng","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf21","name":"Yong Liu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf22","name":"Quanshi Zhang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf23","name":"Lei Zhu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf24","name":"Xingjun Ma","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf25","name":"Junhua Liu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf26","name":"Hui Xue","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf27","name":"Xiaoxiang Zuo","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf28","name":"Xiangnan He","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf29","name":"Chao Shen","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf2a","name":"Xianglong Liu","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf2b","name":"Minlie Huang","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf2c","name":"Jing Shao","hidden":false},{"_id":"6a18fe4056b4bb14ec65cf2d","name":"Xia Hu","hidden":false}],"publishedAt":"2026-05-28T00:00:00.000Z","submittedOnDailyAt":"2026-05-29T00:00:00.000Z","title":"AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security","submittedOnDailyBy":{"_id":"6621f4eb64e84619e578aad6","avatarUrl":"/avatars/b1ad96ee354b999fcafb2998a636609c.svg","isPro":false,"fullname":"Dongrui Liu","user":"shenqiorient","type":"user","name":"shenqiorient"},"summary":"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.","upvotes":104,"discussionId":"6a18fe4056b4bb14ec65cf2e","projectPage":"https://ai45lab.github.io/AgentDoG/v1_5/","ai_summary":"A lightweight and scalable agent safety alignment framework is proposed to address emerging threats from advanced AI models, featuring taxonomy-guided training with minimal samples and efficient deployment in real-world scenarios.","ai_keywords":["agent safety alignment framework","agent safety taxonomy","influence-function purification","AgentDoG 1.5","agentic safety SFT","RL training environment","Docker-level environments","online guardrail","real-time safety moderation","interactive agentic scenarios"],"organization":{"_id":"6747ee5decec679eafb90450","name":"ShanghaiAiLab","fullname":"shanghai ailab ","avatar":"https://www.gravatar.com/avatar/6cd2acf412ad103653d9ce14a1aacc19?d=retro&size=100"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6621f4eb64e84619e578aad6","avatarUrl":"/avatars/b1ad96ee354b999fcafb2998a636609c.svg","isPro":false,"fullname":"Dongrui Liu","user":"shenqiorient","type":"user"},{"_id":"68b4286d2511563aba883bf4","avatarUrl":"/avatars/694b76207ed8483c297bd1427d724761.svg","isPro":false,"fullname":"Qihao Lin","user":"lqh201106","type":"user"},{"_id":"6763f633e90f0e38c79a69cf","avatarUrl":"/avatars/125c3f93229a3a794863723ac0fa9088.svg","isPro":false,"fullname":"Frank Chen","user":"quantumfr","type":"user"},{"_id":"664c85f7dbdb31053d91681d","avatarUrl":"/avatars/029c44aefb90c0c56c139bb8cd151ae5.svg","isPro":false,"fullname":"Qinghua Mao","user":"maopopovich","type":"user"},{"_id":"682844d0f300562abd28e0c9","avatarUrl":"/avatars/8cd5e929cfa19331f38a6f9c97f841b0.svg","isPro":false,"fullname":"dsadsa","user":"yueyue0407","type":"user"},{"_id":"643dfd235aafbdca3a5792c0","avatarUrl":"/avatars/ce8553cf5936012c692e08054ee27937.svg","isPro":false,"fullname":"Bo Zhang","user":"BoZhang","type":"user"},{"_id":"64741e806a972f252de629cc","avatarUrl":"/avatars/597e83bed563aea9d830da90ac1baa68.svg","isPro":false,"fullname":"SII-Qiaosheng Zhang","user":"EricZhang7851","type":"user"},{"_id":"69bbf6dc278d849bc2e5a01b","avatarUrl":"/avatars/69fe25940779230dca7fcc3a5f5e2d84.svg","isPro":false,"fullname":"tian wang","user":"tianjiao24","type":"user"},{"_id":"64cb54bcbb5d195b99186e15","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/64cb54bcbb5d195b99186e15/Z7ziJXjGLXN5rBEYC0sqx.png","isPro":false,"fullname":"SII-Xingjun Ma","user":"xingjunm","type":"user"},{"_id":"67d18a7c312ed7eef068feb9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/z6RVAfS8IZ7SZWUcNvR-v.png","isPro":false,"fullname":"QuWanying","user":"RainbowQTT","type":"user"},{"_id":"6659fe615c3b2634b78572ec","avatarUrl":"/avatars/104ac6a514be366ad6dae71985c769ad.svg","isPro":false,"fullname":"PsychoXiong","user":"PsychoO","type":"user"},{"_id":"667ec2ffa016ced375edfae9","avatarUrl":"/avatars/73f1298569fb0667194d1ac17fb508d6.svg","isPro":false,"fullname":"Ziyun dai","user":"Oliviadzy","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":1,"organization":{"_id":"6747ee5decec679eafb90450","name":"ShanghaiAiLab","fullname":"shanghai ailab ","avatar":"https://www.gravatar.com/avatar/6cd2acf412ad103653d9ce14a1aacc19?d=retro&size=100"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.29801.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.29801 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2605.29801 in a Space README.md to link it from this page.