AI-Augmented Email Analysis Spots Latest Scams, Bad Content

This report summarizes Sophos research and industry commentary on how multimodal AI and LLMs can be used to both create highly personalized, automated phishing/scam campaigns (a so-called “scampaign” using text, image, audio, UI and advertising agents) and improve detection by defenders using multimodal classification (achieving high F1 scores for unseen brand impersonation). It highlights benefits for attackers (scale, microtargeting, faster innovation), defensive gains (reduced false positives, contextual detection), and practical limits such as computational cost and increased data/training needs, while noting widespread real-world misuse has not yet been observed.