Abstract
Phishing has remained one of the central vectors of cyber compromise despite notable progress in the design of secure communication platforms, user-authentication frameworks and email-filtering technologies. Over the last decade, attackers have shifted from repetitive template-driven messages to highly adaptive, context-sensitive campaigns capable of circumventing static filtering rules. This review examines the conceptual and technological evolution of anti-phishing systems through four stages: deterministic rule sets, statistical filters, classical machine-learning classifiers and modern NLP-driven architectures. The analysis focuses on how linguistic interpretation, link-intelligence modelling and behavioural scoring became the structural foundation of contemporary detection pipelines. Emerging research trends are integrated throughout the discussion to illustrate how defence strategies adapt to changes in the threat landscape.