AI-Based Detection of Dark Patterns in Web Interfaces

Dark patterns are deceptive interface designs that influence users to make decisions they may not otherwise choose, such as accepting unwanted tracking, purchasing additional products, or facing difficulties while cancelling subscriptions. Traditional manual auditing and single-technique detection methods may fail to identify deceptive cues distributed across webpage text, visual layouts, and interaction context. Artificial Intelligence (AI) offers potential solutions through Machine Learning (ML), Natural Language Processing (NLP), Computer Vision (CV), and multimodal analysis. This paper reviews selected research on AI-based dark pattern detection and the susceptibility of Large Language Model (LLM)-based web agents to deceptive interfaces. Based on these research directions, a conceptual framework is proposed that combines webpage data collection, feature extraction, textual and visual analysis, pattern classification, explanation generation, and human review. Existing approaches are compared, and challenges related to dataset quality, generalisation, false predictions, explainability, and computational cost are discussed. The proposed framework is conceptual and has not been implemented or experimentally validated. Future work should evaluate its effectiveness using diverse datasets and reproducible experiments.