Metraipolis › archive › dc_997a70ef290a

BabelSteering: Multilingual Safety Alignment via English Steering Vectors — arXiv:2608.16577

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.16577
License (as archived)
CC0-1.0
Archived
2026-10-18T19:01:01.153Z
SHA-256 of the archived text
19797572d8e263cf95622bcf7181b3f9d2858225a1e4a8133845d7941606b8d0
Stable URL of this page
https://metraipolis.com/documento/dc_997a70ef290a

How to cite

BabelSteering: Multilingual Safety Alignment via English Steering Vectors — arXiv:2608.16577 — arXiv (en), https://arxiv.org/abs/2608.16577 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_997a70ef290a (Metraipolis archive dc_997a70ef290a, sha256 19797572d8e263cf95622bcf7181b3f9d2858225a1e4a8133845d7941606b8d0)
BibTeX
@misc{dc997a70ef290a,
  title = {BabelSteering: Multilingual Safety Alignment via English Steering Vectors — arXiv:2608.16577},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.16577},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_997a70ef290a; sha256 19797572d8e263cf95622bcf7181b3f9d2858225a1e4a8133845d7941606b8d0},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.16577. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models (LLMs) are deployed globally in high-stakes settings, yet most safety research and alignment efforts remain concentrated on English. Thus, users interacting with LLMs in other languages may encounter weaker safeguards despite relying on the same systems for similarly sensitive tasks. In this work, we investigate whether safety signals learned from a high-resource language, like English, can improve multilingual safety. We propose BabelSteering, an activation steering method that acts as a lightweight inference- time intervention, using refusal directions derived from English safety supervision to generalize across languages. Our evaluation includes eight languages and jointly measures refusal of harmful requests, over-refusal, and general task utility. The results show that BabelSteering increases the refusal of harmful requests across languages, with only a marginal to no reduction in task utility but with some increase in refusal of pseudo-harmful prompts. For example, for Gemma 7B, we see an average increase in the refusal of harmful prompts across languages of 11 percentage points (pp), with individual languages like Bengali seeing an increase of 17 pp, with no loss of utility on Global MMLU, while pseudo-harmful refusals increase by 13 pp on average. We also introduce a multilingual translation-and-evaluation pipeline to facilitate future work on cross-lingual safety interventions. Overall, our findings suggest that activation steering may provide a practical, low- cost mechanism for extending English-derived safety signals to other languages. Warning: this paper contains examples with unsafe content

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-18 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.16577 — 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 2026-10-18T18:06:57.900Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-08-18?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).