What’s Your Secret?: Secret Scanning by DeepPass2

The report introduces DeepPass2, a multi-tier secret-scanning system that integrates NoseyParker regex rules, a fine-tuned xlm-RoBERTa model for contextual password extraction, and LLM validation to reduce false positives; it details synthetic and augmented dataset creation, chunked-document training, and performance results (e.g., 86.67% strict and 97.72% overlap accuracy) demonstrating robust detection of both structured tokens and free-form passwords, with deployment via Flask and planned Nemesis integration.