Metraipolis › archive › dc_cf5c4bb69953
Not the Same Protector: Deployment-Dependent Protective Intervention in LLMs — arXiv:2608.29136
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.29136
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T07:47:31.155Z
- SHA-256 of the archived text
001b405816c4640a65e7bbd13f99bfc4e0448a59d30a36a51b61386d67a6fdb1- Stable URL of this page
https://metraipolis.com/documento/dc_cf5c4bb69953
How to cite
Not the Same Protector: Deployment-Dependent Protective Intervention in LLMs — arXiv:2608.29136 — arXiv (en), https://arxiv.org/abs/2608.29136 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cf5c4bb69953 (Metraipolis archive dc_cf5c4bb69953, sha256 001b405816c4640a65e7bbd13f99bfc4e0448a59d30a36a51b61386d67a6fdb1)
BibTeX
@misc{dccf5c4bb69953,
title = {Not the Same Protector: Deployment-Dependent Protective Intervention in LLMs — arXiv:2608.29136},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.29136},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cf5c4bb69953; sha256 001b405816c4640a65e7bbd13f99bfc4e0448a59d30a36a51b61386d67a6fdb1},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.29136. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
We ask whether a model protects a user in the same way when that user speaks rather than types. Using a single distress vignette---a physical injury of unstated severity following an interpersonal conflict---we present four frontier models with matched inputs across voice, text, and raw API deployment conditions (n=30 per cell) and code each response along five binary protective indicators, including whether the model issues an explicit medical-care directive. Voice-interface responses are markedly shorter than text-interface responses for three of the four models, and protective behavior contracts alongside that compression: medical directives are at ceiling under both the API and text conditions but decline under voice for every model tested. The contraction is not reducible to length. One model produces voice and text responses of comparable length yet still drops medical directives, and another falls below ceiling between its API and voice conditions, whose responses are of nearly identical length. Under raw API access the pattern is categorical rather than partial: no model asks after the user's safety even once. These results show that protective intervention is sensitive to the surface through which a request arrives, that this sensitivity is detectable using a simple protective coding scheme, and that it is not explained by turn length alone. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.AI). Categorie arXiv: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2608.29136 — 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 2026-10-17T07:28:26.720Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).