From Sports to Safety: Benchmarking Proactive Risk Inference in MLLMs

2026-08-05

AI Summary

This work addresses the limited capability of existing multimodal large language models (MLLMs) to proactively predict physical hazards in real-world safety applications. To this end, we introduce SPRINT, a novel benchmark that leverages sports scenarios as a testbed for active risk reasoning, built upon 2,888 real-world videos spanning 14 sports and three environmental settings, with fine-grained accident annotations and matched safe-control samples. We systematically evaluate MLLMs through multimodal modeling, spatiotemporal cue analysis, hierarchical attribution labeling, and diverse prompting strategies to assess their hazard detection and causal reasoning abilities. Our results reveal that while top-performing models achieve over 95% accuracy in identifying danger signals, their performance drops below 50% in pinpointing underlying causes. Moreover, explicit queries frequently trigger false alarms on safe videos, indicating that current MLLMs exhibit only superficial vigilance without deep causal understanding.

Abstract

Timely anticipation of physical hazards is essential for real-world safety, yet existing MLLM evaluations focus on harmful content or general risks, leaving proactive physical hazard prediction underexplored. Sports provide a well-suited testbed: accident causes span diverse injury dimensions and pre-accident spatiotemporal cues draw on reasoning capabilities shared with broader safety domains such as autonomous driving and fall detection. We introduce SPRINT (Sports Proactive Risk INference Testbed), a benchmark of 2,888 real-world sports videos (2,440 accident, 448 safe controls) spanning 14 sports and 3 environmental settings. Accident videos feature fine-grained annotations of early hazard cues, accident timing, and hierarchical causes; safe videos are manually verified as accident-free and serve to diagnose prompt-induced false alarms. Evaluating state-of-the-art MLLMs under diverse prompts and temporal windows reveals a sharp gap between hazard sensitivity and understanding: the best model exceeds 95% in signaling hazards yet falls below 50% in identifying their causes. Diagnostic experiments further show that explicit danger queries trigger severe false alarms even on hazard-free videos. These findings indicate that current MLLMs exhibit only superficial proactive safety, lacking stable, cause-grounded early warning, and underscore the need for reliable proactive safety in dynamic physical environments. Data and code will be open-sourced upon acceptance.

Problem

• proactive risk inference
• physical hazard prediction
• multimodal large language models
• early warning
• safety benchmarking

Innovation

• proactive risk inference
• multimodal large language models
• sports safety benchmark
• hazard anticipation
• causal reasoning