Metraipolis › archive › dc_d8be2b993ffa

Multimodal Safety Evaluation Should Measure Controllability Beyond Classification — arXiv:2610.06452

Source: arXiv
Original: https://arxiv.org/abs/2610.06452
License (as archived): CC0-1.0
Archived: 2026-10-06T05:25:35.187Z
SHA-256 of the archived text: b6227492b53b660bf2b4e5d9c0e1f39a2c27574fa29e1082bd9716f717cc5a66
Stable URL of this page: https://metraipolis.com/documento/dc_d8be2b993ffa

How to cite

Multimodal Safety Evaluation Should Measure Controllability Beyond Classification — arXiv:2610.06452 — arXiv, https://arxiv.org/abs/2610.06452 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_d8be2b993ffa (Metraipolis archive dc_d8be2b993ffa, sha256 b6227492b53b660bf2b4e5d9c0e1f39a2c27574fa29e1082bd9716f717cc5a66)

BibTeX

@misc{dcd8be2b993ffa,
  title = {Multimodal Safety Evaluation Should Measure Controllability Beyond Classification — arXiv:2610.06452},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.06452},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_d8be2b993ffa; sha256 b6227492b53b660bf2b4e5d9c0e1f39a2c27574fa29e1082bd9716f717cc5a66},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.06452. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

VLM safety is commonly evaluated through input- and output-level classification. Such classification is necessary, but it does not reveal whether a safety state is accessible or controllable inside the model. We argue that multimodal safety evaluation should therefore report a \emph{controllability profile} alongside behavioral classification, separating representation-level detectability, cross-modal specificity, intervention sensitivity, and benign-preserving selectivity. Using implicit toxicity as a stress case, we instantiate this profile on LlavaGuard and Qwen3.5 with sparse feature decompositions. LlavaGuard admits localized handles with a narrow benign-preserving intervention range and modest downstream safety gains, whereas Qwen3.5 supports strong representation-level readout but no comparable selective-control regime under the tested operators. These results show that internal readout and controllability can diverge. Future multimodal safety benchmarks should therefore report not only behavioral safety metrics, but also whether safety-relevant internal signals can be intervention-tested and controlled within a validated operating range.

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/2610.06452 — 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 2026-10-06T05:24:05.678Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).