PassGPT

PassGPT is a specialized GPT-2 language model trained exclusively on password leak datasets. It advances password security research by outperforming previous generative methods and enabling guided generation with arbitrary constraints.

Key Features

• Superior Password Guessing — Outperforms GAN-based methods by guessing twice as many previously unseen passwords.
• Guided Password Generation — Adapts sampling to generate passwords matching specific constraints, a capability lacking in GAN approaches.
• Password Strength Estimation — Uses model-assigned probabilities to enhance existing password strength estimators.
• Character-Level Tokenization — Preserves meaningful probability distributions by avoiding letter concatenation into single tokens.
• Research-Focused Design — Optimized for non-commercial research use with curated pre-trained models available.

Philosophy

PassGPT is designed as a research tool to advance password security through large language models, with careful attention to preserving meaningful probability distributions and enabling controlled generation scenarios.