AI-Based Multimodal Phishing Detection Using Large Language Models

Phishing is one of the most common cybersecurity threats where attackers trick users through fake emails, websites, or links to steal sensitive information. With the rapid development of artificial intelligence, phishing attacks are becoming more realistic and harder to detect using traditional methods. Earlier systems relied on blacklists, rules, and URL features, but recent research explores deep learning, Large Language Models (LLMs), and multimodal approaches. Explainability and privacy-aware techniques such as federated learning are also gaining importance. The increasing sophistication of phishing campaigns highlights the need for advanced detection frameworks that can analyze diverse information sources simultaneously. Mul-timodal systems leverage text, URLs, visual webpage content, and embedded scripts to provide a more comprehensive evaluation of suspicious activity. Large Language Mod-els further enhance semantic understanding, enabling detection of subtle manipulations in phishing messages. By combining multimodal analysis with explainable and privacy-preserving AI, researchers aim to build adaptive, transparent, and trustworthy solutions that can effectively counter evolving phishing threats in real-world cybersecurity environ-ments.