Targeted Password Guessing Using k-Nearest Neighbors

We propose KNNGuess/KNN-TPG as a new targeted password guessing model based on old password. We show how to use our trained model (with the trained model weights in ./experiment and the trained KNN-TPG database in ./datastore) to evaluate the strength of a current password based on an old one (see Figure 12 in the paper). Some of the data used to generate the experimental plots is shown in ./exp_ans.

Experimental Setup:

• Microarchitecture

  • CPU Model: Intel Xeon Silver processor
  • OS: Ubuntu 20.04 LTS
  • Linux Kernel: 5.11.0-46-generic
  • GPU: NVIDIA RTX 3090 GPU
  • RAM: 256GB

• Environment

  • Python 3.7
  • torch 1.13.0+cu116
  • torchvision 0.14.1+cu116
  • word2keypress 1.0.16
  • faiss-gpu 1.7.3

You can use it to build environment:

conda install pytorch torchvision torchaudio pytorch-cuda=11.6 -c pytorch -c nvidia
conda install -c pytorch -c nvidia fass-gpu=1.7.3

Run

Evaluating password strength

python psm.py --source_password YOUR_OLD_PASSWORD --target_password YOUR_NEW_PASSWORD

“YOUR_NEW_PASSWORD” is the password whose strength needs to be evaluated. A detailed description of the KNN-PSM can be found in Appendix H of the paper.

The output is the probability of each character in YOUR_NEW_PASSWORD being predicted by KNNGuess. The greater the probability, the less secure the character is, and the easier it is for the password to be guessed by an attacker.