---
id: 20260627-T0-04
title: "研究发现：AI 拒绝回答受限于“人设”特征"
title_en: "Research Finds Model Refusal is Gated by Persona Traits"
url: https://ai.daily.yangsir.net/daily/20260627-T0-04
issue_date: 2026-06-27
publish_date: 2026-06-26T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2606.26161
---

# 研究发现：AI 拒绝回答受限于“人设”特征

arXiv 新研究揭示了聊天模型中“拒绝机制”与“人设机制”的相互作用。研究指出，模型激活空间中的线性方向控制着拒绝和顺从行为，且一个“顺从的人设”会充当守门员的角色，直接阻碍模型触发拒绝行为。这意味着调整模型的性格特征可能会改变其安全防御的稳定性。

## English Version

**Research Finds Model Refusal is Gated by Persona Traits**

New research on arXiv (cs.AI) reveals that the refusal mechanism in instruction-tuned chat models is gated by persona traits. The study identifies linear directions in activation space for both refusal and persona, showing that a compliant persona can actively inhibit refusal behaviors. This interaction suggests that persona adjustments can impact the model's safety alignment.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2606.26161)

**详情页**：https://ai.daily.yangsir.net/daily/20260627-T0-04

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*