Search papers, labs, and topics across Lattice.
The paper introduces SpecularNet, a reference-free web phishing detection framework that uses a hierarchical graph autoencoder to model the DOM tree structure of webpages. SpecularNet employs directional, level-wise message passing to capture structural invariants indicative of phishing attempts, operating solely on domain names and HTML structure. Experimental results demonstrate that SpecularNet achieves a 93.9% F1 score on benchmark datasets with significantly reduced inference time (20ms per webpage) compared to reference-based methods, while also showing robustness in real-world deployments and against adversarial attacks.
A lightweight, reference-free architecture rivals heavyweight phishing detectors, achieving near state-of-the-art performance with orders of magnitude faster inference.
Phishing remains the most pervasive threat to the Web, enabling large-scale credential theft and financial fraud through deceptive webpages. While recent reference-based and generative-AI-driven phishing detectors achieve strong accuracy, their reliance on external knowledge bases, cloud services, and complex multimodal pipelines fundamentally limits practicality, scalability, and reproducibility. In contrast, conventional deep learning approaches often fail to generalize to evolving phishing campaigns. We introduce SpecularNet, a novel lightweight framework for reference-free web phishing detection that demonstrates how carefully designed compact architectures can rival heavyweight systems. SpecularNet operates solely on the domain name and HTML structure, modeling the Document Object Model (DOM) as a tree and leveraging a hierarchical graph autoencoding architecture with directional, level-wise message passing. This design captures higher-order structural invariants of phishing webpages while enabling fast, end-to-end inference on standard CPUs. Extensive evaluation against 13 state of the art phishing detectors, including leading reference-based systems, shows that SpecularNet achieves competitive detection performance with dramatically lower computational cost. On benchmark datasets, it reaches an F1 score of 93.9%, trailing the best reference-based method slightly while reducing inference time from several seconds to approximately 20 milliseconds per webpage. Field and robustness evaluations further validate SpecularNet in real-world deployments, on a newly collected 2026 open-world dataset, and against adversarial attacks.
