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

arXiv:2508.11290v3 Announce Type: replace Abstract: 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-refusal rates by up to 73% with minimal impact on utility -- offering a principled and conditional approach to mitigating over-refusals.

Related

• What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal
• Guardian-as-an-Advisor: Advancing Next-Generation Guardian Models for Trustworthy LLMs
• Deliberative Alignment is Deep, but Uncertainty Remains: Inference time safety improvement in reasoning via attribution of unsafe behavior to base model
• Jailbreaking the Matrix: Nullspace Steering for Controlled Model Subversion
• Self-Debias: Self-correcting for Debiasing Large Language Models

Source: arXiv cs.CL | 2026-04-14