Metraipolis › archive › dc_a97b315fa6df
When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs — arXiv:2608.29936
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.29936
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T08:07:16.489Z
- SHA-256 of the archived text
97870ab6fd3229d629e4589ba559760339d8f2348d5d23ca8a79eca71cc95011- Stable URL of this page
https://metraipolis.com/documento/dc_a97b315fa6df
How to cite
When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs — arXiv:2608.29936 — arXiv (en), https://arxiv.org/abs/2608.29936 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a97b315fa6df (Metraipolis archive dc_a97b315fa6df, sha256 97870ab6fd3229d629e4589ba559760339d8f2348d5d23ca8a79eca71cc95011)
BibTeX
@misc{dca97b315fa6df,
title = {When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMs — arXiv:2608.29936},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.29936},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a97b315fa6df; sha256 97870ab6fd3229d629e4589ba559760339d8f2348d5d23ca8a79eca71cc95011},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.29936. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Safety alignment of large language models (LLMs) degrades across languages, yet the internal mechanism driving this asymmetry remains poorly understood. Our work, therefore, presents a systematic mechanistic analysis of multilingual safety using sparse autoencoder (SAE) features, sparse interpretable directions in the residual stream associated with harmful and harmless model behavior across three instruction-tuned LLMs, eight languages, and all model layers. We observe that safety-relevant features are architecture-dependent in terms of where they are located and how they are distributed across layers. Additionally, they are geometrically entangled with language identity and exhibit cross-lingual sharing patterns, i.e., languages share safety features to varying degrees across model depths and architectures. This safety-language entanglement has direct consequences such that ablating safety features impacts not only harmful response rates but also target language, with the degree of intervention predicted by the relationship between safety and language features. Our findings qualify the language-universality of safety alignment as architecture-dependent and offer a mechanistic account of multilingual safety interventions. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2608.29936 — 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 2026-10-17T07:47:52.758Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).