Metraipolis › archive › dc_7aed48a4ebbe

On the Steering Dimensionality of Refusal in Language Models — arXiv:2610.04245

Source: arXiv
Original: https://arxiv.org/abs/2610.04245
License (as archived): CC0-1.0
Archived: 2026-10-06T05:24:24.635Z
SHA-256 of the archived text: 0330a4a6f22fd6cc2f111ca39078f572841e2a16d6bebc2cd8b5700426a41b46
Stable URL of this page: https://metraipolis.com/documento/dc_7aed48a4ebbe

How to cite

On the Steering Dimensionality of Refusal in Language Models — arXiv:2610.04245 — arXiv, https://arxiv.org/abs/2610.04245 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7aed48a4ebbe (Metraipolis archive dc_7aed48a4ebbe, sha256 0330a4a6f22fd6cc2f111ca39078f572841e2a16d6bebc2cd8b5700426a41b46)

BibTeX

@misc{dc7aed48a4ebbe,
  title = {On the Steering Dimensionality of Refusal in Language Models — arXiv:2610.04245},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.04245},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7aed48a4ebbe; sha256 0330a4a6f22fd6cc2f111ca39078f572841e2a16d6bebc2cd8b5700426a41b46},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.04245. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Existing activation steering methods often assume that a high-level concept can be mediated by a single steering direction. To support this, two complementary interventions should be achieved: additive steering should induce the target behavior, while directional ablation should suppress it. Yet behaviors may occupy richer activation geometries beyond a single direction, and semantically similar behaviors may be represented by distinct directions. In this work, we study how many directions can reliably control two different types of refusal behaviors: refusal triggered by the safety alignment and refusal in general contexts. Given the limited expressive capability of a single steering vector, we study the general setting of steering subspaces and introduce the notion of steering dimensionality as the minimum subspace dimensionality required to reliably control a behavior. We characterize sufficient steering subspaces that cover the full extent of the target behavior through both the (monotonic) improvement before the sufficient dimensionality, and the saturation beyond it. Empirically, we find that refusal triggered by safety alignment is 1-dim steerable, while multiple distinct steering directions can achieve comparable control. In contrast, refusal in general contexts exhibits substantially richer activation geometry where even 5-dim steering subspaces fail to reliably capture its full steerable variation. Our results reveal that the activation geometry underlying refusal is highly context-dependent and can be substantially more complex than a single linear steering direction.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.04245 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-06T05:24:05.445Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).