A Hybrid Deep-Learning Framework for Detecting AI-Generated Phishing Attacks in Cloud-Based Communication Systems

Abstract

Generative artificial intelligence has altered the operational character of phishing. Large language models can produce fluent, context-sensitive and highly personalised messages that resemble routine organisational correspondence, weakening filters that depend heavily on spelling errors, suspicious keywords or previously observed templates. This conceptual study develops a hybrid deep-learning framework for identifying both conventional and AI-assisted phishing in cloud-based communication systems. Following a design-science approach, the paper synthesises current standards, threat reports and recent research on transformer-based classification, adversarially modified phishing and explainable artificial intelligence. The proposed architecture combines six evidence streams: contextual semantics, stylometry, URL and domain properties, authentication headers, sender behaviour and attachment risk. An attention-based fusion layer produces a calibrated risk estimate, while an explanation layer communicates the principal factors behind each decision. The framework does not assume that AI-generated text is inherently malicious; instead, it assesses deceptive intent, technical inconsistency and contextual abnormality. The study argues that robust protection requires multi-source evidence, continuous monitoring, human oversight and privacy-sensitive deployment. Keywords: artificial intelligence; phishing detection; cloud security; transformers; behavioural analytics; explainable AI; adversarial robustness