CVE-2026-31214 describes an insecure deserialization vulnerability in the torch-checkpoint-shrink.py script of the ml-engineering project. The script uses torch.load() to deserialize PyTorch checkpoint files (.pt) without enabling the security-restrictive weights_only=True parameter. This omission permits deserialization of arbitrary Python objects via pickle, enabling remote attackers to execute arbitrary code by supplying malicious checkpoint files.

Successful exploitation allows arbitrary code execution with the privileges of the user running the vulnerable script. This can lead to full system compromise or unauthorized actions within the user's environment. There are no known exploits in the wild at this time.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, users should avoid loading untrusted PyTorch checkpoint files with torch.load() without the weights_only=True parameter. Applying this parameter restricts deserialization to tensor weights only, mitigating the risk of arbitrary code execution.