Understanding and Mitigating Overrefusal in LLMs from an Unveiling Perspective of Safety Decision Boundary

2025-05-23 · Licheng Pan, Yongqi Tong, Xin Zhang, Xiaolu Zhang, Jun Zhou, Zhixuan Chu

Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks, yet they often refuse to answer legitimate queries-a phenomenon known as overrefusal. Overrefusal typically stems from over-conservative safety alignment, causing models to treat many reasonable prompts as potentially risky. To systematically understand this issue, we probe and leverage the models'safety decision boundaries to analyze and mitigate overrefusal. Our findings reveal that overrefusal is closely tied to misalignment at these boundary regions, where models struggle to distinguish subtle differences between benign and harmful content. Building on these insights, we present RASS, an automated framework for prompt generation and selection that strategically targets overrefusal prompts near the safety boundary. By harnessing steering vectors in the representation space, RASS efficiently identifies and curates boundary-aligned prompts, enabling more effective and targeted mitigation of overrefusal. This approach not only provides a more precise and interpretable view of model safety decisions but also seamlessly extends to multilingual scenarios.We have explored the safety decision boundaries of various LLMs and construct the MORBench evaluation set to facilitate robust assessment of model safety and helpfulness across multiple languages. Code and datasets will be released at https://anonymous.4open.science/r/RASS-80D3.

📄 PDF Abstract BibTeX arXiv:2505.18325

Code (0)

등록된 구현이 없습니다.

Tasks

Safety Alignment

Methods 이 논문이 사용한 방법론

SET Dynamic Sparse Training method where weight mask is updated randomly periodically

Understanding and Mitigating Over-refusal for Large Language Models via Representation Intervention

2025-11-24 · Junbo Zhang, Ran Chen, Qianli Zhou, Xinyang Deng 외 arxiv

Large language models (LLMs) demonstrate powerful capabilities across various natural language processing tasks,yet their inherent safety vulnerabilities undermine the reliable application of LLMs in real-world scenarios…

Deactivating Refusal Triggers: Understanding and Mitigating Overrefusal in Safety Alignment

2026-03-12 · Zhiyu Xue, Zimo Qi, Guangliang Liu, Bocheng Chen 외 arxiv

Safety alignment aims to ensure that large language models (LLMs) refuse harmful requests by post-training on harmful queries paired with refusal answers. Although safety alignment is widely adopted in industry, the over…

DDOR: Delta Debugging for Explainable Overrefusal Testing and Repair

2026-06-02 · Qinyan Zhou, Peixin Zhang, Jun Sun, Haonan Zhang 외 arxiv

While safety alignment and guardrails help large language models (LLMs) avoid harmful outputs, they can also induce overrefusal, i.e., unwarranted rejection of benign queries that merely appear risky. We present DDOR (De…

Verb Mirage: Unveiling and Assessing Verb Concept Hallucinations in Multimodal Large Language Models

2024-12-06 · Zehao Wang, Xinpeng Liu, Xiaoqian Wu, Yudonglin Zhang 외

Multimodal Large Language Models (MLLMs) have garnered significant attention recently and demonstrate outstanding capabilities in various tasks such as OCR, VQA, captioning, $\textit{etc}$. However, hallucination remains…

Hallucination Optical Character Recognition (OCR)Visual Question Answering (VQA)

POROver: Improving Safety and Reducing Overrefusal in Large Language Models with Overgeneration and Preference Optimization

2024-10-16 · Batuhan K. Karaman, Ishmam Zabir, Alon Benhaim, Vishrav Chaudhary 외

Balancing safety and usefulness in large language models has become a critical challenge in recent years. Models often exhibit unsafe behavior or adopt an overly cautious approach, leading to frequent overrefusal of beni…

Instruction Following