EVADE-Bench: Multimodal Benchmark for Evasive Content Detection in E-Commerce Applications

TL;DR

EVADE-Bench introduces a comprehensive multimodal benchmark dataset designed to evaluate and improve the ability of foundation models to detect evasive, misleading content in e-commerce, addressing a critical challenge in online content moderation.

Contribution

This paper presents the first expert-curated, multimodal benchmark specifically targeting evasive content detection in e-commerce, including new tasks, datasets, and baseline evaluations for foundation models.

Findings

State-of-the-art models often misclassify evasive samples.

Clearer rule definitions improve model alignment and reasoning.

Benchmark exposes fundamental limitations in current multimodal models.

Abstract

E-commerce platforms increasingly rely on Large Language Models (LLMs) and Vision-Language Models (VLMs) to detect illicit or misleading product content. However, these models remain vulnerable to evasive content: inputs (text or images) that superficially comply with platform policies while covertly conveying prohibited claims. Unlike traditional adversarial attacks that induce overt failures, evasive content exploits ambiguity and context, making it far harder to detect. Existing robustness benchmarks provide little guidance for this demanding, real-world challenge. We introduce EVADE, the first expert-curated, Chinese, multimodal benchmark specifically designed to evaluate foundation models on evasive content detection in e-commerce. The dataset contains 2,833 annotated text samples and 13,961 images spanning six demanding product categories, including body shaping, height growth, and health supplements. Two complementary tasks assess distinct capabilities: Single-Violation, which probes fine-grained reasoning under short prompts, and All-in-One, which tests long-context reasoning by merging overlapping policy rules into unified instructions. Notably, the All-in-One setting significantly narrows the performance gap between partial and full-match accuracy, suggesting that clearer rule definitions improve alignment between human and model judgment. We benchmark 26 mainstream LLMs and VLMs and observe substantial performance gaps: even state-of-the-art models frequently misclassify evasive samples. By releasing EVADE and strong baselines, we provide the first rigorous standard for evaluating evasive-content detection, expose fundamental limitations in current multimodal reasoning, and lay the groundwork for safer and more transparent content moderation systems in e-commerce. The dataset is publicly available at https://huggingface.co/datasets/koenshen/EVADE-Bench.