Combating GenAI Email Attacks with BERT LLM · Blog · Sublime Security

This report describes the growing use of Generative AI to produce highly personalized and scalable phishing campaigns (notably BEC and vendor impersonation), explains why detecting whether text is AI-generated is less useful than identifying malicious intent, and outlines Sublime Security’s layered detection approach using Message Query Language (MQL), NLU models (fine-tuned BERT/DistilBERT), data augmentation with synthetic samples, and behavioral/contextual signals to improve detection and reduce false positives.