The vulnerability in PyTorch-Lightning (≤ 2.6.0) arises from the load_from_checkpoint() method calling torch.load() without the security-restrictive parameter weights_only=True. This omission allows deserialization of arbitrary Python objects via Pickle, leading to potential arbitrary code execution if a malicious checkpoint file is loaded. The issue is a classic insecure deserialization flaw (CWE-502) in the model checkpoint loading process. No CVSS score or official remediation is currently published.

An attacker able to provide a malicious checkpoint file to a victim loading it with load_from_checkpoint() can execute arbitrary code on the victim's system. This could lead to full system compromise depending on the environment and privileges of the process loading the checkpoint. The vulnerability is local to the environment loading the checkpoint and requires attacker control or influence over the checkpoint file source.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, users should avoid loading checkpoint files from untrusted or unauthenticated sources. Consider manually inspecting checkpoint files or using alternative safe loading mechanisms if possible.