CVE Catalog

CVE-2026-79784

HighCVSS 8.8
Published: Updated: Translated: NVD NIST

Exploitation Probability (EPSS)

Low risk
0.32%

25th percentile - higher than 25% of all known CVEs

Summary

The Vocos library instantiates a class named by a configuration file without restricting which class may be named. The instantiate_class function imports the module and calls the specified attribute with arguments from the config, allowing arbitrary code execution. Loading a model from a Hugging Face repository not controlled by the user can execute code chosen by the repository owner.

Risk Assessment

A user loading a model from an untrusted repository may execute malicious code during the loading process, leading to system compromise or data theft.

Recommendation

Update the Vocos library to a version that restricts allowed classes in the configuration (allowlist). Until then, load models only from trusted repositories.

Related vulnerabilities

Original NVD description (English source)

Vocos instantiates a class named by a configuration file without restricting which class may be named. instantiate_class in vocos/pretrained.py takes the class_path value from the configuration, splits it into a module and an attribute, imports the module with __import__, resolves the attribute with getattr, and calls the result as args_class(*args, **kwargs) where kwargs is the config's own init_args mapping. No allowlist constrains the dotted path, so a configuration may name any importable callable and supply the arguments it is called with. Vocos.from_hparams reaches this for each of the feature_extractor, backbone and head entries, and Vocos.from_pretrained reaches it with a remote file: it downloads config.yaml from a caller-named Hugging Face repository and passes it straight to from_hparams. Loading a model from a repository the user does not control therefore executes code of the repository owner's choosing in the loading process. The neighbouring torch.load of the downloaded weights is a separate matter and is constrained on PyTorch releases that default weights_only to true, which leaves this path as the reachable one.

Vulnerability data from NVD (NIST) · CISA KEV · EPSS