There Is More to Refusal in Large Language Models than a Single Direction
This work addresses the problem of understanding and controlling refusal mechanisms in large language models for AI safety and alignment researchers, providing a more nuanced view than previous incremental findings.
The paper challenges the prior claim that refusal in large language models is mediated by a single activation-space direction, showing instead that refusal behaviors correspond to geometrically distinct directions across eleven categories, but linear steering along any refusal-related direction produces similar refusal-to-over-refusal trade-offs as a shared control knob.
Prior work argues that refusal in large language models is mediated by a single activation-space direction, enabling effective steering and ablation. We show that this account is incomplete. Across eleven categories of refusal and non-compliance, including safety, incomplete or unsupported requests, anthropomorphization, and over-refusal, we find that these refusal behaviors correspond to geometrically distinct directions in activation space. Yet despite this diversity, linear steering along any refusal-related direction produces nearly identical refusal to over-refusal trade-offs, acting as a shared one-dimensional control knob. The primary effect of different directions is not whether the model refuses, but how it refuses.