tensorflow-1.4-billion-password-analysis

Why we included this project

This project feeds a large dump of real-world leaked passwords through a character-level TensorFlow model and uses what it learns to study how people actually create and change them. Security researchers, penetration testers, and anyone building password-strength or credential-hygiene tooling get a complete, reproducible pipeline: scripts to process the corpus, a training run, and edit-distance analysis of password variants that shows the kind of mutation users make, like hello123 turning into h@llo123 and then h@llo!23. The processing scripts let you run a smaller sample before committing to the full 50 GB and several hours of training, which keeps the cost of exploration low. It is a research starting point and reference implementation, clearly marked for research purposes only, not a production service.