PT-2026-30793 · Hugging Face · Huggingface/Transformers

Published 2026-04-07 · Updated 2026-04-07 · CVE-2026-1839

NVD

CVSS v3.1

6.5

Medium

AV:L/AC:H/PR:N/UI:R/S:U/C:H/I:L/A:H

A vulnerability in the HuggingFace Transformers library, specifically in the Trainer class, allows for arbitrary code execution. The load rng state() method in src/transformers/trainer.py at line 3059 calls torch.load() without the weights only=True parameter. This issue affects all versions of the library supporting torch>=2.2 when used with PyTorch versions below 2.6, as the safe globals() context manager provides no protection in these versions. An attacker can exploit this vulnerability by supplying a malicious checkpoint file, such as rng state.pth, which can execute arbitrary code when loaded. The issue is resolved in version v5.0.0rc3.

Fix

Deserialization of Untrusted Data

Weakness Enumeration

CWE-502

Related Identifiers

CVE-2026-1839

Affected Products

Huggingface/Transformers

References · 3

• https://nvd.nist.gov/vuln/detail/CVE-2026-1839 · Security Note
• https://github.com/huggingface/transformers/commit/03c8082ba4594c9b8d6fe190ca9bed0e5f8ca396 · Note
• https://huntr.com/bounties/3c77bb97-e493-493d-9a88-c57f5c536485 · Note