HarmVideoBench: Benchmarking Harmful Video Understanding in Large Multimodal Models

2026-06-25

AI Summary

Existing benchmarks for evaluating harmful videos predominantly rely on binary classification, lacking fine-grained measurement and interpretability for implicit and deep semantic harms. To address this limitation, this work proposes HarmVideoBench—a multi-level diagnostic benchmark comprising 1,379 videos and 4,137 multiple-choice questions—that introduces a novel three-tier progressive evaluation framework encompassing observable evidence, intra-clip semantics, and extra-clip reasoning. Interpretability is achieved through explanatory question answering, enabling transparent model assessment. Additionally, the authors propose a Budget-Constrained Retrieval (BCR) method that dynamically retrieves contextual information on demand to enhance multimodal reasoning. Experiments across 19 state-of-the-art multimodal large language models demonstrate that BCR significantly improves macro-average accuracy from 61.7% to 84.4%, establishing a new state-of-the-art performance.

Abstract

Large vision-language models (LVLMs) have recently shown immense potential in automated content moderation, sparking growing interest in developing harmful-video benchmarks. However, we identify two primary limitations in existing works: 1) The multi-layered characteristics of harmful videos are overlooked. Existing benchmarks predominantly formulate evaluation as a binary classification task, failing to capture implicit or deep contextual harms. 2) Explanatory rationales are completely absent. Current frameworks measure exclusively whether a model flags a video correctly rather than explaining why, turning evaluation into a black box where models can succeed through superficial shortcuts. To address these problems, we present HarmVideoBench, a multi-layered diagnostic benchmark comprising 1,379 videos paired with 4,137 multiple-choice questions. HarmVideoBench benchmarks three hierarchical dimensions: Observable Evidence, Clip-Internal Meaning, and Beyond-Clip Reasoning, aiming to evaluate models' deep understanding beyond surface cues with carefully balanced and curated samples. We evaluate 19 leading models on HarmVideoBench to assess their multidimensional understanding of harmful videos. Moreover, we introduce BCR, a benchmark-aligned method that predicts reasoning boundaries and dynamically retrieves context only when needed. Experimental results show that BCR substantially improves the base model's performance in harmful video understanding, raising the macro average from 61.7 percent to a state-of-the-art 84.4 percent.

Problem

• harmful video understanding
• multimodal benchmark
• explainable evaluation
• deep contextual harm
• content moderation

Innovation

• HarmVideoBench
• multilayered benchmark
• beyond-clip reasoning
• explainable evaluation
• context-aware retrieval