PHISHGUARD AI: A Hybrid Multi-Modal Phishing Detection System Using Transformer Intelligence, Domain Analytics, and Behavioral Verification

Abstract. Phishing has become a critical cybersecurity threat, where attackers imper sonate trusted entities through fraudulent emails and websites to steal sensitive infor mation. Conventional detection approaches based on static blacklists and keyword filtering are often ineffective against zero-day and highly obfuscated attacks. This research pro poses a multi-modal phishing detection framework that integrates deep learning, heuris tic analysis, and behavioral monitoring. A transformer-based attention model analyzes se mantic patterns in URLs and email content, while domain intelligence derived from WHOIS metadata evaluates domain legitimacy. Additionally, a sandboxed behavioral crawler identifies hidden forms, malicious scripts, and suspicious redirects. A hybrid scor ing mechanism combines these signals to generate an interpretable safety score. Experi mental evaluation demonstrates improved detection accuracy 0.98% and reduced false positives, with supporting real-time deployment in browsers and email systems.