Email Topic Modeling: Simplifying detection with ML-powered granularity · Blog · Sublime Security

Beta announcement for Topic Modeling: a machine-learning feature that classifies email content into 27 predefined topics with confidence levels and attachment/OCR support. The team used few-shot LLM prompting to generate labeled training data, then trained a faster NLU-style classifier for production use. Topic Modeling is intended to improve spam/graymail verdicts and simplify behavioral detections via MQL (example: credential-phishing detection) and is available in beta for custom detection rules, threat hunts, and automations.