The rise of generative artificial intelligence (AI) has significantly increased the sophistication of phishing attacks, allowing malicious actors to craft highly convincing, context-aware emails that can evade traditional detection systems. This paper proposes a novel framework for detecting and preventing generative AI-driven phishing by leveraging digital twin technology. Digital twins—virtual replicas of users and email systems—model and monitor communication and behavioural patterns to identify anomalies indicative of phishing attempts. By integrating natural language processing (NLP), machine learning (ML), and anomaly detection, the framework achieved a detection accuracy of 97.8%, with precision of 98.1% and recall of 96.7%. It successfully detected over 92% of generative AI-crafted phishing emails, including highly tailored and context-aware attacks. This research contributes by introducing a multi-faceted detection approach, combining BERT, RNN, and Isolation Forest algorithms to address linguistic, behavioural, and metadata-based anomalies. The framework’s ability to model dynamic user behaviours with digital twin technology enhances its adaptability to evolving threats. The findings highlight the potential for scalable, high-performance phishing detection, offering a robust solution to safeguard organisations against AI-driven cyberattacks. Future work will explore multimodal attack detection and computational optimization for large-scale deployments.