FraudBench: A Multimodal Benchmark for Detecting AI-Generated Fraudulent Refund Evidence

2026-05-09

AI Summary

This study addresses the growing misuse of AI-generated synthetic images to fabricate evidence of product damage for fraudulent refund claims—a problem exacerbated by the inability of existing detection methods to effectively correlate visual content with accompanying claim narratives. To tackle this challenge, the work introduces a novel multimodal verification framework that jointly analyzes textual descriptions, images, and metadata within the specific context of refund claims. The authors also present FraudBench, a benchmark dataset spanning e-commerce, food delivery, and travel services, comprising real–forged image pairs where synthetic damages are generated from authentic undamaged images using six state-of-the-art generative models. Experimental results reveal that current multimodal large language models (MLLMs) achieve fraud detection rates below 50%, while specialized detectors, though more effective, still suffer from limited generalization across generative models and high false-positive rates on genuinely damaged items.

Abstract

Artificial Intelligence (AI)-generated images have become increasingly realistic and readily adaptable to concrete real-world claims, creating new challenges for verifying visual evidence. A concrete emerging risk is AI-generated refund fraud, in which manipulated or synthetic images are used to support claims about damaged products, poor delivery conditions, or service-related defects. Existing AI-generated image detection benchmarks mainly evaluate standalone authenticity classification, cross-generator transfer, or forensic localization, leaving claim-conditioned fraudulent evidence detection underexplored. To bridge this gap, we introduce FraudBench, a multimodal benchmark for detecting AI-generated fraudulent refund evidence. FraudBench is constructed from real-world user-review evidence across e-commerce, food delivery, and travel-service scenarios. We curate real evidence images together with their associated review and product metadata, identify genuine damaged and undamaged evidence through MLLM-assisted filtering and human annotation, and synthesize fake-damaged evidence from genuine undamaged reference images using six state-of-the-art image editing and generation models. Using FraudBench, we evaluate MLLMs, specialized AI-generated image detectors, and human participants under the same settings. Experiments show that current MLLMs often recognize real-damaged evidence but fail on many fake-damaged subsets, with fake-damage detection rates (TPR) far below the 50% baseline on most generator subsets. Specialized detectors generally perform better but remain inconsistent across generators and can produce false positives on real-damaged samples, revealing a clear gap between generic AI image detection and reliable claim-conditioned refund-evidence verification.

Problem

• AI-generated fraud
• refund evidence
• multimodal benchmark
• claim-conditioned verification
• synthetic image detection

Innovation

• multimodal benchmark
• AI-generated fraud detection
• claim-conditioned verification
• synthetic evidence
• MLLM evaluation