\n <img src=\"https://cdn-uploads.huggingface.co/production/uploads/662df118b0d404635d7a2b46/0whXCio2sdCw94a-EUbjx.png\" width=\"85%\" alt=\"GateMem\">\n</p>\n\nGateMem is a benchmark for memory governance in multi-principal shared-memory agents. Instead of only asking whether an agent can remember information, GateMem evaluates whether a persistent-memory agent can remain useful, enforce requester-specific access control, and honor deletion requests.\n\n<p align=\"center\">\n <img src=\"https://cdn-uploads.huggingface.co/production/uploads/662df118b0d404635d7a2b46/XZgiRXYpu4MZiQfThVq1h.png\" width=\"95%\" alt=\"GateMem pl\">\n</p>\n\nWe release:\n* 91 long-form multi-party episodes\n* 2,218 hidden evaluation checkpoints\n* 4 shared-memory domains: medical, office, education, and household\n* 7 memory-agent baselines across 6 backbone LLMs\n* official evaluation code\n* a public leaderboard and online submission interface\n\nResources:\n* 📄 Paper: https://arxiv.org/abs/2606.18829\n* 🌐 Project page: https://rzhub.github.io/GateMem/project.html\n* 💻 Code: https://github.com/rzhub/GateMem\n* 🤗 Dataset: https://huggingface.co/datasets/Ray368/GateMem\n* 🏆 Leaderboard: https://rzhub.github.io/GateMem/\n* 🚀 Submit results: https://huggingface.co/spaces/Ray368/GateMem-Submit\n\nWe welcome new submissions to the leaderboard and feedback from the community.","html":"<p>Thanks for checking out GateMem!</p>\n<p align=\"center\">\n <img src=\"https://cdn-uploads.huggingface.co/production/uploads/662df118b0d404635d7a2b46/0whXCio2sdCw94a-EUbjx.png\" width=\"85%\" alt=\"GateMem\">\n</p>\n\n<p>GateMem is a benchmark for memory governance in multi-principal shared-memory agents. Instead of only asking whether an agent can remember information, GateMem evaluates whether a persistent-memory agent can remain useful, enforce requester-specific access control, and honor deletion requests.</p>\n<p align=\"center\">\n <img src=\"https://cdn-uploads.huggingface.co/production/uploads/662df118b0d404635d7a2b46/XZgiRXYpu4MZiQfThVq1h.png\" width=\"95%\" alt=\"GateMem pl\">\n</p>\n\n<p>We release:</p>\n<ul>\n<li>91 long-form multi-party episodes</li>\n<li>2,218 hidden evaluation checkpoints</li>\n<li>4 shared-memory domains: medical, office, education, and household</li>\n<li>7 memory-agent baselines across 6 backbone LLMs</li>\n<li>official evaluation code</li>\n<li>a public leaderboard and online submission interface</li>\n</ul>\n<p>Resources:</p>\n<ul>\n<li>📄 Paper: <a href=\"https://arxiv.org/abs/2606.18829\" rel=\"nofollow\">https://arxiv.org/abs/2606.18829</a></li>\n<li>🌐 Project page: <a href=\"https://rzhub.github.io/GateMem/project.html\" rel=\"nofollow\">https://rzhub.github.io/GateMem/project.html</a></li>\n<li>💻 Code: <a href=\"https://github.com/rzhub/GateMem\" rel=\"nofollow\">https://github.com/rzhub/GateMem</a></li>\n<li>🤗 Dataset: <a href=\"https://huggingface.co/datasets/Ray368/GateMem\">https://huggingface.co/datasets/Ray368/GateMem</a></li>\n<li>🏆 Leaderboard: <a href=\"https://rzhub.github.io/GateMem/\" rel=\"nofollow\">https://rzhub.github.io/GateMem/</a></li>\n<li>🚀 Submit results: <a href=\"https://huggingface.co/spaces/Ray368/GateMem-Submit\">https://huggingface.co/spaces/Ray368/GateMem-Submit</a></li>\n</ul>\n<p>We welcome new submissions to the leaderboard and feedback from the community.</p>\n","updatedAt":"2026-06-22T05:11:54.321Z","author":{"_id":"662df118b0d404635d7a2b46","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662df118b0d404635d7a2b46/JtcwRctghymSkw67zvoPn.jpeg","fullname":"Zhe Ren","name":"Ray368","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":6,"identifiedLanguage":{"language":"en","probability":0.729583203792572},"editors":["Ray368"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/662df118b0d404635d7a2b46/JtcwRctghymSkw67zvoPn.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.18829","authors":[{"_id":"6a33e7b7fc3a8b1102d94397","user":{"_id":"662df118b0d404635d7a2b46","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662df118b0d404635d7a2b46/JtcwRctghymSkw67zvoPn.jpeg","isPro":false,"fullname":"Zhe Ren","user":"Ray368","type":"user","name":"Ray368"},"name":"Zhe Ren","status":"claimed_verified","statusLastChangedAt":"2026-06-19T14:20:52.358Z","hidden":false},{"_id":"6a33e7b7fc3a8b1102d94398","name":"Yibo Yang","hidden":false},{"_id":"6a33e7b7fc3a8b1102d94399","name":"Yimeng Chen","hidden":false},{"_id":"6a33e7b7fc3a8b1102d9439a","name":"Zijun Zhao","hidden":false},{"_id":"6a33e7b7fc3a8b1102d9439b","name":"Benshuo Fu","hidden":false},{"_id":"6a33e7b7fc3a8b1102d9439c","name":"Zhihao Shu","hidden":false},{"_id":"6a33e7b7fc3a8b1102d9439d","name":"Bingjie Zhang","hidden":false},{"_id":"6a33e7b7fc3a8b1102d9439e","name":"Yangyang Xu","hidden":false},{"_id":"6a33e7b7fc3a8b1102d9439f","name":"Dandan Guo","hidden":false},{"_id":"6a33e7b7fc3a8b1102d943a0","name":"Shuicheng Yan","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/662df118b0d404635d7a2b46/Pc05fy1hRuYm06ceFfNT2.jpeg","https://cdn-uploads.huggingface.co/production/uploads/662df118b0d404635d7a2b46/FHuKmwoFQxzdYNDha_0Dv.jpeg"],"publishedAt":"2026-06-17T00:00:00.000Z","submittedOnDailyAt":"2026-06-22T00:00:00.000Z","title":"GateMem: Benchmarking Memory Governance in Multi-Principal Shared-Memory Agents","submittedOnDailyBy":{"_id":"662df118b0d404635d7a2b46","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662df118b0d404635d7a2b46/JtcwRctghymSkw67zvoPn.jpeg","isPro":false,"fullname":"Zhe Ren","user":"Ray368","type":"user","name":"Ray368"},"summary":"Memory benchmarks for LLM agents largely assume single-user settings, leaving shared assistants for hospitals, workplaces, campuses, and households understudied. In these deployments, multiple principals write to a common memory pool and query it under different roles, scopes, and relationships, so memory quality requires governance as well as recall. We introduce GateMem, a benchmark for multi-principal shared-memory agents. GateMem jointly evaluates utility for legitimate long-horizon requests with state updates, access control across contextual authorization boundaries, and agent-facing active forgetting after explicit deletion requests. It spans medical, office, education, and household domains, with long-form multi-party episodes, incremental memory injection, hidden checkpoints, structured judging, and leak-target annotations. Across diverse baselines and backbone models, no method simultaneously achieves strong utility, robust access control, and reliable forgetting. Long-context prompting often yields the best governance score at high token cost, while retrieval-based and external-memory methods reduce cost yet still leak unauthorized or deleted information. These results show current memory agents remain far from reliable shared institutional deployment.","upvotes":4,"discussionId":"6a33e7b8fc3a8b1102d943a1","projectPage":"https://rzhub.github.io/GateMem/project.html","githubRepo":"https://github.com/rzhub/GateMem","githubRepoAddedBy":"user","ai_summary":"Current memory agents lack reliable shared institutional deployment due to challenges in balancing utility, access control, and forgetting across multiple principals with diverse authorization contexts.","ai_keywords":["memory benchmarks","multi-principal shared-memory agents","access control","active forgetting","long-context prompting","retrieval-based methods","external-memory methods"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":63},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"680c615cb1024379fac47b8c","avatarUrl":"/avatars/1480a62811146de955a782da9e9c8212.svg","isPro":false,"fullname":"Li tonghui","user":"litonghui","type":"user"},{"_id":"69ddb3390f67b65cbfefca87","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/hBYjOj2a_RK2yeWodleSV.jpeg","isPro":false,"fullname":"付本硕","user":"fubenshuo","type":"user"},{"_id":"662df118b0d404635d7a2b46","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/662df118b0d404635d7a2b46/JtcwRctghymSkw67zvoPn.jpeg","isPro":false,"fullname":"Zhe Ren","user":"Ray368","type":"user"},{"_id":"6a38cad617951799b4a363c9","avatarUrl":"/avatars/4b9f92fe5c28c3c30143cf73ccbec18c.svg","isPro":false,"fullname":"sun fu","user":"sunyanzhu","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":3,"query":{}}">

Cite arxiv.org/abs/2606.18829 in a model README.md to link it from this page.

Add this paper to a collection to link it from this page.