Picklescan Bugs Allow Malicious PyTorch Models to Evade Scans and Execute Code

Picklescan Bugs Allow Malicious PyTorch Models to Evade Scans and Execute Code

Three critical security flaws have been found in Picklescan, an open-source security utility for Python pickle files, which could enable attackers to bypass protections and execute malicious code. These vulnerabilities pose a significant threat to machine learning supply chains, especially when loading untrusted PyTorch models. #Picklescan #PyTorch #SupplyChainAttack

Keypoints

  • Three vulnerabilities in Picklescan allow bypassing its malware detection capabilities.
  • The flaws enable execution of malicious code through crafted pickle files and extensions.
  • Exploits can also disable ZIP archive scanning and undermine dangerous import blocklists.
  • Recent updates have addressed these vulnerabilities in version 0.0.31.
  • Third-party security flaws showcase systemic issues and the need for adaptive AI model security measures.

Read More: https://thehackernews.com/2025/12/picklescan-bugs-allow-malicious-pytorch.html