arXiv cs.AI· Yewon Han, Jisu Jeon, Suhyun Choi, Juan Yeo, Minsoo Jo, Gihoon Kim, Hyunggyu Lim, Taesup Kim·· 3 天前AI 评分38
STOU：第一视角多模态助手安全评测基准，区分伤害对象而非只看拒答率
Safety for Whom? Benchmarking Understanding Beyond Refusal in Egocentric Multimodal Assistants
AI 导读
研究者提出 STOU 基准，包含 2,902 条第一视角视频-音频与视频-文本查询，按潜在伤害对象分为用户自身（Self）、同场他人（The Others）与社会整体（Others）。该评测将安全理解与拒答行为解耦，除拒答外还考察模型是否识别用户意图、伤害的存在与对象及其后果。在 10 个开源与闭源全模态及多模态 LLM 上，拒答率随伤害对象系统性变化，且模型常识别出风险仍继续提供帮助。
正文 · 原文
Abstract:Egocentric multimodal assistants see through the user's eyes and guide what the user does next, so their outputs directly shape physical actions, including those that affect bystanders. Yet existing safety evaluations do not distinguish whom a response would harm, and score only whether a model refuses. Pooling all harms into one refusal rate hides two failures: a model may protect one target but not another, and it may refuse without understanding the risk, or understand it and comply anyway. We present STOU, a benchmark of 2,902 egocentric video-audio and video-text queries whose potential harm targets the user (Self), co-present individuals (The Others), or society at large (Others). Our evaluation decouples safety understanding from refusal behavior: beyond refusal, we score whether the model recognizes the user's intent, the presence and target of harm, and its downstream consequences. Across 10 open and closed source omnimodal and multimodal LLMs, we find that refusal varies systematically with the harm target, and that models often recognize a harm yet still assist with it. These results show that refusal rates alone cannot certify an assistant as safe, and that safety alignment must account for who would be harmed. Project page: this https URL
| Subjects: | Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC) |
| Cite as: | arXiv:2610.17728 [cs.AI] |
| (or arXiv:2610.17728v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17728 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yewon Han [view email]
[v1]
Tue, 13 Oct 2026 05:05:10 UTC (41,105 KB)
来源：arXiv cs.AI · arxiv.org