CVE Catalog

CVE-2026-57845

HighCVSS 8.8
Published: Updated: Translated: NVD NIST

Exploitation Probability (EPSS)

Low risk
0.55%

46th percentile - higher than 46% of all known CVEs

Summary

Hugging Face Datasets through 5.1.0 contains an insecure deserialization vulnerability allowing arbitrary Python code execution via a crafted pickle file. The file is loaded through the deprecated pandas packaged builder and processed by Pandas._generate_tables via pandas.read_pickle() without validation, so calling datasets.load_dataset("pandas", ...) executes the embedded __reduce__ payload with the victim process's privileges.

Risk Assessment

An attacker can achieve remote code execution in the victim's environment, leading to system compromise, data theft, or further escalation. The risk applies to any organization loading untrusted pandas-format datasets.

Recommendation

Upgrade Hugging Face Datasets to a version newer than 5.1.0 and avoid loading untrusted pickle files. Prefer safer data formats where possible and verify dataset sources.

Related vulnerabilities

Original NVD description (English source)

Hugging Face Datasets through 5.1.0 contains an insecure deserialization vulnerability that allows attackers to execute arbitrary Python code by crafting a malicious pickle file loaded through the deprecated pandas packaged builder. Attackers can supply a pickle-backed dataset that is processed by Pandas._generate_tables via pandas.read_pickle() without validation, so when a victim calls datasets.load_dataset("pandas", data_files=...), the embedded __reduce__ payload executes code with the victim process's privileges.

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