From Sports to Safety: Benchmarking Proactive Risk Inference in MLLMs
Jiawei Qiu, Yichen Xu, Jianzhe Ma, Mingyang Yu, Wenbin Zhu, Yang Han, Pinzheng Lv, Wenxuan Wang
cs.CV, cs.CL
Submitted: 2026-08-06
Comments: Preprints
Code: https://github.com/DawnGavial/SPRINT
License: http://creativecommons.org/licenses/by/4.0/
The gist: 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
Terminology
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.
Sources
- STaR-GATE: Teaching Language Models to Ask Clarifying Questions
- HealthBench: Evaluating Large Language Models Towards Improved Human Health
- Qwen3-VL Technical Report
- SafeWatch: An Efficient Safety-Policy Following Video Guardrail Model with Transparent Explanations
- A Survey on Proactive Dialogue Systems: Problems, Methods, and Prospects
- RESPONSE: Benchmarking the Ability of Language Models to Undertake Commonsense Reasoning in Crisis Situation
- Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned
- DisasterResponseGPT: Large Language Models for Accelerated Plan of Action Development in Disaster Response Scenarios
- GPT-4o System Card
- QuestBench: Can LLMs ask the right question to acquire information in reasoning tasks?
- Video-SafetyBench: A Benchmark for Safety Evaluation of Video LVLMs
- Proactive Agent: Shifting LLM Agents from Reactive Responses to Active Assistance
- HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
- Feedback Loops With Language Models Drive In-Context Reward Hacking
- Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!
- Identifying the Risks of LM Agents with an LM-Emulated Sandbox
- Seed1.8 Model Card: Towards Generalized Real-World Agency
- OpenAI GPT-5 System Card
- Gemini: A Family of Highly Capable Multimodal Models
- SafeArena: Evaluating the Safety of Autonomous Web Agents
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models