• Python version: 3.9.13

  conda create -n picklecloak python=3.9.13
  conda activate picklecloak

• Necessary libs: please refer to the README.md under each component folder.

This artifact contains three folders:

• gadget: includes the artifacts for disclosing the risky function surface using static analysis and the LLM-augmented AEG pipeline.
• loading_surface: includes the artifacts for analyzing the model loading surface, consisting of CodeQL queries used to examine pickle loading paths and extract exceptions for Exception-Directed Programming (EDP).
• craft_model: contains code demonstrating how the risky function surface and model loading surface can be leveraged to craft malicious models.

Before exploring the code, please read the corresponding README.md in each folder for details on structure and usage.