RiskCueBench: Benchmarking Anticipatory Reasoning from Early Risk Cues in Video-Language Models
Sha Luo, Yogesh Prabhu, Timothy Ossowski, Kaiping Chen, Junjie Hu · 2026
Read RiskCueBench: Benchmarking Anticipatory Reasoning from Early Risk Cues in Video-Language Models by Sha Luo, Yogesh Prabhu, Timothy Ossowski, Kaiping Chen, Junjie Hu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-03369.txt.
Description
Sha Luo and four co-authors build RiskCueBench, a benchmark that tests whether video-language models can spot danger before it happens. Social media video is full of risky moments, and anticipating them could prevent real accidents. Earlier datasets cheated in effect: models saw the full video, accident included, which makes the task far easier than real forecasting. RiskCueBench instead annotates each video with a risk signal clip, the earliest moment hinting at a possible safety problem.
Models must judge the situation from those early cues alone. Experiments reveal a significant gap in current systems: they struggle to read evolving situations and anticipate risky events from early visual signals. The benchmark is meant to expose that weakness honestly, so progress on deployable risk prediction can be measured.
Catalog record
| CREATOR | Sha Luo, Yogesh Prabhu, Timothy Ossowski, Kaiping Chen, Junjie Hu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-03369