Towards Evaluating Proactive Risk Awareness of Multimodal Language Models

2025-05-23

AI Summary

Human security awareness gaps impede timely risk identification in daily life. To address this, we introduce PaSBench—the first multimodal benchmark for proactive risk perception—comprising 416 cross-modal security scenarios to shift AI from passive response to proactive warning. We propose a novel “observe–reason–warn” evaluation paradigm, jointly modeling image sequences and textual logs, with structured risk annotations across five safety-critical domains and systematic failure attribution analysis. Evaluating 36 state-of-the-art models reveals that top-performing models achieve only 71% (image) and 64% (text) accuracy, with 45–55% of risks consistently missed; the primary bottleneck is unstable proactive reasoning—not knowledge deficiency. Based on these findings, we identify three key research directions: trustworthy proactive reasoning, temporal risk modeling, and cross-modal warning alignment. PaSBench is publicly available on Hugging Face.

Technology Category

• Natural Language Processing: Safety and Robustness
• Knowledge Representation and Reasoning: Common-Sense Reasoning
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Human safety awareness gaps often prevent the timely recognition of everyday risks. In solving this problem, a proactive safety artificial intelligence (AI) system would work better than a reactive one. Instead of just reacting to users' questions, it would actively watch people's behavior and their environment to detect potential dangers in advance. Our Proactive Safety Bench (PaSBench) evaluates this capability through 416 multimodal scenarios (128 image sequences, 288 text logs) spanning 5 safety-critical domains. Evaluation of 36 advanced models reveals fundamental limitations: Top performers like Gemini-2.5-pro achieve 71% image and 64% text accuracy, but miss 45-55% risks in repeated trials. Through failure analysis, we identify unstable proactive reasoning rather than knowledge deficits as the primary limitation. This work establishes (1) a proactive safety benchmark, (2) systematic evidence of model limitations, and (3) critical directions for developing reliable protective AI. We believe our dataset and findings can promote the development of safer AI assistants that actively prevent harm rather than merely respond to requests. Our dataset can be found at https://huggingface.co/datasets/Youliang/PaSBench.

Problem

• Evaluating proactive risk awareness in multimodal language models
• Assessing AI's ability to detect potential dangers in advance
• Identifying limitations in proactive reasoning for safety-critical scenarios

Innovation

• Proactive AI detects risks in behavior and environment
• Multimodal evaluation with 416 diverse safety scenarios
• Identifies unstable reasoning as main model limitation