Generative AI–Enabled Phishing: Emerging Attack Capabilities, Detection Limitations, and Research Directions for Adaptive Cybersecurity

Generative artificial intelligence (AI) is changing how phishing messages are written and adapted. Large language models can produce clear, convincing, and varied messages, making some traditional warning signs less reliable. These tools may also help security teams examine suspicious emails and support decisions. This paper reviews research on AI-enabled phishing, focusing on phishing emails, message generation, detection methods, and limits in current evaluation practices. It compares approaches that use message content, context, machine learning, and large language models to identify phishing attempts. The review finds that strong results on one dataset do not prove that a method will work against new campaigns or in different settings. Data quality, false alarms, missed attacks, privacy, cost, and clear explanations also matter. The paper identifies research gaps and suggests future priorities, including testing across time and sources, combining useful signals, and updating detection systems carefully. The proposed framework needs further testing and validation