安全为谁？拒绝话题中的正确子集而非整个话题
原标题：Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic
AI 摘要
Hugging Face Blog 介绍了一篇新论文《Safety for Whom? Boundary-Aware Self-Distillation for Controlled LLM Safety Refusal》，研究如何让模型只拒绝某个话题中与部署政策不兼容的子集，而非整个话题。作者以政治说服为测试场景，指出标准自生成训练流程存在覆盖缺口、误拒副作用和边界度量缺失三个问题，并提出升级重试、分布内良性数据和边界配对等修复方法。实验显示，在 Qwen3-8B 上政治拒绝率从 9.47% 升至 84.75%，但 XSTest 过度拒绝率也从 2.00% 升至 74.00%，说明必须同时报告安全与过度拒绝两个维度。
正文节选
Real deployments rarely fit the topic-level picture. The same base model may be adapted for a general assistant, an educational product, an enterprise system, or a public-sector service, and each setting needs different boundaries within the same topic. A civics tutor and a public-sector assistant can share a model yet require opposite behaviour on politics: both should answer factual questions about an election, but only one may need to refuse a request to write targeted political manipulation.