Fraudulent shopping websites pose a significant threat to online consumers and legitimate businesses: in 2023, victims of such scams reported $392 million in losses to the Federal Trade Commission. This alarming trend not only impacts individuals but also erodes societal trust in e-commerce, necessitating urgent countermeasures. While previous studies have attempted to identify these fraudulent websites at scale, they face limitations such as potential bias in data collection, overreliance on easily manipulated features, and the lack of explainable results. This study explores the potential of Large Language Models (LLMs) in identifying fraudulent shopping websites, revealing that current LLMs underperform compared to existing machine learning models. To address this, we propose ScamNet, a fine-tuned LLM for explainable fraudulent shopping website detection. Our experimental results on real-world datasets demonstrate a breakthrough in detection performance from 22.35% detection rate to 95.59%, particularly in identifying subtle deceptive tactics such as using a legitimate-looking website template. ScamNet offers interpretable insights into its decision-making process, enhancing transparency and overcoming a key limitation of previous approaches.

Despite recent advancements in malicious website detection and phishing mitigation, the security ecosystem has paid little attention to Fraudulent e-Commerce Websites (FCWs), such as fraudulent shopping websites, fake charities, and cryptocurrency scam websites. Even worse, there are no active large-scale mitigation systems or publicly available datasets for FCWs.In this paper, we first propose an efficient and automated approach to gather FCWs through crowdsourcing. We identify eight different types of non-phishing FCWs and derive key defining characteristics. Then, we find that anti-phishing mitigation systems, such as Google Safe Browsing, have a detection rate of just 0.46% on our dataset. We create a classifier, BEYOND PHISH, to identify FCWs using manually defined features based on our analysis. Validating BEYOND PHISH on never-before-seen (untrained and untested data) through a user study indicates that our system has a high detection rate and a low false positive rate of 98.34% and 1.34%, respectively. Lastly, we collaborated with a major Internet security company, Palo Alto Networks, as well as a major financial services provider, to evaluate our classifier on manually labeled real-world data. The model achieves a false positive rate of 2.46% and a 94.88% detection rate, showing potential for real-world defense against FCWs.