There Is More to Refusal in Large Language Models than a Single Direction
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866914301561798656 |
|---|---|
| author | Joad, Faaiz Hawasly, Majd Boughorbel, Sabri Durrani, Nadir Sencar, Husrev Taha |
| author_facet | Joad, Faaiz Hawasly, Majd Boughorbel, Sabri Durrani, Nadir Sencar, Husrev Taha |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_02132 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | There Is More to Refusal in Large Language Models than a Single Direction Joad, Faaiz Hawasly, Majd Boughorbel, Sabri Durrani, Nadir Sencar, Husrev Taha Computation and Language 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. |
| title | There Is More to Refusal in Large Language Models than a Single Direction |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2602.02132 |