GENERATIVE AI FRAMEWORK FOR DETECTING/PREVENTING PASSWORD CRACKING

ABSTRACT Password cracking remains a significant cybersecurity threat, necessitating advanced detection mechanisms to safeguard user authentication systems. This study presents a generative AI framework for detecting and preventing password cracking attempts, leveraging anomaly detection within the CRISP-DM methodology. Using a sampled subset of the RockYou dataset (100,000 passwords) and a synthetic login dataset (10,000 records), an Isolation Forest model was trained to identify anomalous login behaviors indicative of brute-force or dictionary attacks. The system, implemented as a Flask-based web application, achieved Accuracy: 0.90, Precision: 0.90, F1-Score: 0.79, and ROC-AUC: 0.88 on the testing set, demonstrating robust performance for a proof-of-concept. Deployed locally with Docker for portability, the framework offers scalability potential through cloud integration. Ethical considerations, including avoiding raw password storage, were prioritized. While effective, limitations include reliance on synthetic data and the model’s simplicity compared to advanced generative AI techniques. The study provides a lightweight, accessible solution for password security, with recommendations for incorporating GANs and real-world data in future iterations.