Metraipolis › archive › dc_0f305cb1471e
GuardianBench: A Same-Scene Instruction-Contrastive Benchmark for Latent Contextual Risk in Embodied AI — arXiv:2608.21928
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.21928
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-18T07:39:06.452Z
- SHA-256 of the archived text
2b1767c22e09071251238812a1beb085c46194feaee997a0061762dc9db29b05- Stable URL of this page
https://metraipolis.com/documento/dc_0f305cb1471e
How to cite
GuardianBench: A Same-Scene Instruction-Contrastive Benchmark for Latent Contextual Risk in Embodied AI — arXiv:2608.21928 — arXiv (en), https://arxiv.org/abs/2608.21928 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0f305cb1471e (Metraipolis archive dc_0f305cb1471e, sha256 2b1767c22e09071251238812a1beb085c46194feaee997a0061762dc9db29b05)
BibTeX
@misc{dc0f305cb1471e,
title = {GuardianBench: A Same-Scene Instruction-Contrastive Benchmark for Latent Contextual Risk in Embodied AI — arXiv:2608.21928},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.21928},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0f305cb1471e; sha256 2b1767c22e09071251238812a1beb085c46194feaee997a0061762dc9db29b05},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.21928. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
In embodied AI, safety risk can be latent: a benign instruction and a safe scene become hazardous only when composed. Prior work has advanced embodied safety by varying visual contexts or evaluating execution-time dynamics, but the complementary axis of fixing the scene and varying only the instruction remains underexplored. We introduce GuardianBench, an instruction-contrastive benchmark grounded in international safety standards that isolates this latent contextual risk through 3,024 instruction-scene examples organized as same-scene Safe/Unsafe contrastive pairs across various hazard categories. Benchmarking state-of-the-art vision-language models (VLMs) reveals instruction-insensitive verdicts: models disproportionately approve both instructions under a given scene; across the primary models, average pair accuracy is only 24.1%. Our systematic rationale audit localizes the dominant failure: models fail to bind the instruction-relevant cues that differentiate safe from unsafe compositions. As a post-training case study, Verdict Log-Odds Supervision (VLOS), a lightweight verdict-level objective, substantially improves performance on open-weight backbones. Together, our latent contextual risk task formulation, standards-grounded contrastive benchmark construction, pair-level and rationale-level failure diagnosis, and benchmark-enabled verdict calibration establish GuardianBench as a controlled evaluation suite for exposing and improving safety reasoning over instruction-scene compositions under latent contextual risk. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-25 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Robotics (cs.RO). Pagina del lavoro: https://arxiv.org/abs/2608.21928 — 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 2026-10-18T07:37:52.682Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).