analiza aiqbench
GuardianBench introduces a critical evaluation framework for embodied AI, exposing how current vision-language models fail to detect latent safety risks when a benign scene is paired with a hazardous instruction. By organizing thousands of same-scene contrastive pairs, the benchmark reveals that state-of-the-art models often approve both safe and unsafe instructions indiscriminately, highlighting a fundamental gap in binding instruction-relevant cues to contextual hazards. The study further demonstrates that lightweight supervision methods like Verdict Log-Odds Supervision can significantly improve safety reasoning without requiring massive architectural changes.
GuardianBench: A Same-Scene Instruction-Contrastive Benchmark for Latent Contextual Risk in Embodied AI
arXiv:2608.21928v1 Announce Type: new Abstract: 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.
Abstract: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.
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Robotics (cs.RO)
arXiv:2608.21928 [cs.AI]
(or arXiv:2608.21928v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2608.21928 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 287 cuvinte
Subiecte conexe
[AINews] Claude Opus 5.5, the new default model for AINews — and everybody cuts prices 40-50%
Latent Space · 23 sept. 2026
[AINews] not much happened today
Latent Space · 3 oct. 2026
[AINews] not much happened today
Latent Space · 10 sept. 2026
[AINews] High Taste Testers Ep. 1
Latent Space · 17 oct. 2026