The torch-checkpoint-shrink.py script fails to set the security‑restrictive weights_only=True parameter when calling torch.load(), allowing arbitrary Python objects to be deserialized via pickle. This flaw is classified as CWE‑502 (Insecure Deserialization). An attacker who can supply a maliciously crafted .pt checkpoint file can trigger the script to execute arbitrary code in the environment of the user running the script, leading to complete compromise of that system.

The vulnerability exists in the open‑source ml‑engineering project, specifically the torch-checkpoint-shrink.py script. Any installation of this script that is capable of processing external checkpoint files is affected; there are no vendor or product version details provided beyond the open‑source repository identifier.

With no EPSS data and not listed in CISA KEV, the risk is determined solely by the inherent severity of arbitrary code execution. The denial of the weights_only option creates a high‑severity vector; if a user runs the script with a malicious checkpoint, execution can occur with the user’s permissions. The unlikely requirement that the attacker must execute the script or supply a file derived from a trusted source means the vulnerability is mainly exploitable in controlled or insider scenarios, but the impact remains severe.