paper-with-me

Papers

SafeConstellations: Mitigating Over-Refusals in LLMs Through Task-Aware Representation Steering

2025-08-15 · Utsav Maskey, Sumit Yadav, Mark Dras, Usman Naseem arxiv

LLMs increasingly exhibit over-refusal behavior, where safety mechanisms cause models to reject benign instructions that seemingly resemble harmful content. This phenomenon diminishes utility in production applications that repeatedly rely on common prompt templates or applications that frequently rely on LLMs for specific tasks (e.g. sentiment analysis, language translation). Through extensive evaluation, we demonstrate that LLMs persist in refusing inputs containing harmful content, even when they are reframed with tasks that have benign intent. Our mechanistic analysis reveals that LLMs follow distinct "constellation" patterns in embedding space as representations traverse layers, with each NLP task maintaining consistent trajectories that shift predictably between refusal and non-refusal cases. We introduce SafeConstellations, an inference-time trajectory-shifting approach that tracks task-specific trajectory patterns and guides representations toward non-refusal pathways. By selectively guiding model behavior only on tasks prone to over-refusal, our method reduces over-refusals with minimal impact on utility -- offering a principled and conditional approach to mitigating over-refusals.

📄 PDF Abstract BibTeX arXiv:2508.11290

Code (0)

등록된 구현이 없습니다.

Tasks

Sentiment Analysis

Similar Papers 제목 키워드 기반

Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning

2025-07-06 · Mahavir Dabas, Si Chen, Charles Fleming, Ming Jin 외

Safety alignment is crucial for large language models (LLMs) to resist malicious instructions but often results in over-refusals, where benign prompts are unnecessarily rejected, impairing user experience and model utili…

Safety Alignment

Beyond Over-Refusal: Scenario-Based Diagnostics and Post-Hoc Mitigation for Exaggerated Refusals in LLMs

2025-10-09 · Shuzhou Yuan, Ercong Nie, Yinuo Sun, Chenxuan Zhao 외 arxiv

Large language models (LLMs) frequently produce false refusals, declining benign requests that contain terms resembling unsafe queries. We address this challenge by introducing two comprehensive benchmarks: the Exaggerat…

EVOREFUSE: Evolutionary Prompt Optimization for Evaluation and Mitigation of LLM Over-Refusal to Pseudo-Malicious Instructions

2025-05-29 · Xiaorui Wu, Xiaofeng Mao, Fei Li, Xin Zhang 외

Large language models (LLMs) frequently refuse to respond to pseudo-malicious instructions: semantically harmless input queries triggering unnecessary LLM refusals due to conservative safety alignment, significantly impa…

Safety Alignment

FalseReject: A Resource for Improving Contextual Safety and Mitigating Over-Refusals in LLMs via Structured Reasoning

2025-05-12 · Zhehao Zhang, Weijie Xu, Fanyou Wu, Chandan K. Reddy

Safety alignment approaches in large language models (LLMs) often lead to the over-refusal of benign queries, significantly diminishing their utility in sensitive scenarios. To address this challenge, we introduce FalseR…

16kBenchmarkingSafety Alignment

Steering Over-refusals Towards Safety in Retrieval Augmented Generation

2025-10-12 · Utsav Maskey, Mark Dras, Usman Naseem arxiv

Safety alignment in large language models (LLMs) induces over-refusals -- where LLMs decline benign requests due to aggressive safety filters. We analyze this phenomenon in retrieval-augmented generation (RAG), where bot…