The torch-checkpoint-shrink.py script in the ml-engineering project uses torch.load() to deserialize PyTorch checkpoint files (.pt) without enabling the weights_only=True parameter, which restricts deserialization to tensor weights only. This insecure deserialization vulnerability (CWE-502) allows an attacker to supply a malicious checkpoint file that executes arbitrary Python code upon loading. The vulnerability has a CVSS 3.1 base score of 9.8, indicating critical severity with network attack vector, no privileges required, and no user interaction needed.

Successful exploitation allows remote attackers to execute arbitrary code on the system running the vulnerable script, potentially leading to full system compromise. The impact includes complete confidentiality, integrity, and availability loss as indicated by the CVSS vector (C:H/I:H/A:H). No known exploits in the wild have been reported to date.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, users should avoid loading untrusted PyTorch checkpoint files with this script or modify the script to use torch.load() with the weights_only=True parameter to restrict deserialization to tensor weights only.