arXiv cs.AI· Xinyang Liao, Lingyu Li, Kaicheng Shen, Yang Yao, Na Lu, Yan Teng, Xia Hu·· 3 天前AI 评分39
CompanionSafe：从多轮安全评估到 AI 伴侣护栏
CompanionSafe: From Multi-turn Safety Evaluation to Guardrails for AI Companions
AI 导读
研究团队提出 CompanionSafe 统一框架，包含多轮评估组件 CompanionEval 与护栏组件 CompanionGuard。CompanionEval 基于多样人设、匹配伴侣角色和对话场景构建 1300 个关系化多轮评估设置，覆盖四个轮次层级；CompanionGuard 将基准暴露的失败转化为轮次定位监督，用于伴侣特定风险检测。
正文 · 原文
Abstract:AI companions provide users with emotional support and ongoing companionship, but their sustained relational interactions also introduce distinct safety risks. Prior work largely targets general harms or specific companion platforms and risk categories, while relying on manually collected conversations, limiting scalable and user-grounded evaluation and mitigation. To address this, we introduce CompanionSafe, a unified evaluation-to-mitigation framework for AI companion safety. Its evaluation component, CompanionEval, provides scalable multi-turn evaluation by constructing 1,300 relationally grounded settings from diverse personas, matched companion characters, and dialogue scenarios across four turn tiers. Its mitigation component, CompanionGuard, converts benchmark-exposed failures into turn-localized supervision for companion-specific risk detection. Our results show that even strong LLMs and existing guard models remain vulnerable to companion-specific relational harms, while both closed- and open-source models exhibit increasing safety risks as interactions become longer. Together, CompanionEval and CompanionGuard form a practical evaluation-to-mitigation loop for AI companion safety, turning relational safety failures exposed during multi-turn interaction into supervision for companion-specific guardrails.
| Comments: | 40 pages, 33 figures, Code and dataset are available at this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Human-Computer Interaction (cs.HC) |
| Cite as: | arXiv:2610.18095 [cs.AI] |
| (or arXiv:2610.18095v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.18095 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xinyang Liao [view email]
[v1]
Tue, 13 Oct 2026 10:13:07 UTC (3,859 KB)
来源：arXiv cs.AI · arxiv.org