This vulnerability arises from improper validation of user‑supplied checkpoint data in Hugging Face PyTorch Image Models. Deserialization of an untrusted checkpoint can lead to arbitrary code execution in the context of the current process, classified as CWE‑502. The impact is that a remote attacker, using a malicious checkpoint, can run code with the privileges of the process that loads the model.

All installations of Hugging Face PyTorch Image Models are affected. Version details are not provided in the advisory, so any current release should be considered vulnerable until a patch is applied.

The CVSS score of 7.8 indicates a high severity. No EPSS score is available, and the vulnerability is not listed in the CISA KEV catalog, suggesting limited publicly documented exploitation. However, the attack requires user interaction—such as visiting a page or opening a malicious file—which means the attacker can only compromise a target after social engineering or file delivery. Given these constraints, the overall risk remains high but exploitability is confined to contexts where an untrusted checkpoint is loaded.