# CVE-2026-1839: HuggingFace Transformers allows for arbitrary code execution in the `Trainer` class

Source: https://junglewise.ai/threats/cve-2026-1839-huggingface-transformers-allows-for-arbitrary-code-execution-in  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-04-07  
License: CC BY 4.0 with a link back

CVE-2026-1839 · Severity: high · CVSS 7.8 · Published 2026-04-07

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.

Technologies: [Huggingface Transformers](https://junglewise.ai/threats/technologies/huggingface-transformers), [transformers (PyPI)](https://junglewise.ai/threats/technologies/pypi-transformers). Vendors: [Huggingface](https://junglewise.ai/threats/vendors/huggingface), [PyPI](https://junglewise.ai/threats/vendors/pypi).

## Quick answers

**Is CVE-2026-1839 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-1839 being exploited in the wild.

**How severe is CVE-2026-1839?** High severity, with a CVSS score of 7.8, published 7 April 2026.

**Which products and versions are affected?** transformers (PyPI) ECOSYSTEM:introduced=0,fixed=5.0.0, ECOSYSTEM:introduced=0,fixed=5.0.0rc3.

**Is there a fix for CVE-2026-1839?** Yes. transformers (PyPI): upgrade to 5.0.0, 5.0.0rc3 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

**Am I affected by CVE-2026-1839?** Check whether you run Huggingface Transformers or transformers (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## How to fix

- transformers (PyPI): upgrade to 5.0.0, 5.0.0rc3 or later

  ```sh
  pip install "transformers>=5.0.0rc3"
  ```


## Affected products

- Huggingface Transformers
- transformers (PyPI) ECOSYSTEM:introduced=0,fixed=5.0.0, ECOSYSTEM:introduced=0,fixed=5.0.0rc3

## Identifiers

- CVE-2026-1839
- GHSA-69w3-r845-3855
- PYSEC-2026-2288

## Related threats

- [Hugging Face Transformers arbitrary file write via load_custom_generate](https://junglewise.ai/threats/cve-2026-80047-hugging-face-transformers-arbitrary-file-write-via-load-custom)
- [Hugging Face Transformers path traversal in save_pretrained](https://junglewise.ai/threats/cve-2026-9856-hugging-face-transformers-path-traversal-in-save-pretrained)
- [Hugging Face Transformers vulnerable to Regular Expression Denial of Service (ReDoS) in the AdamWeightDecay optimizer](https://junglewise.ai/threats/cve-2025-6921-hugging-face-transformers-vulnerable-to-regular-expression-denial)
- [PYSEC-2026-1988 - Hugging Face Transformers library has Regular Expression Denial of Service](https://junglewise.ai/threats/cve-2025-6051-hugging-face-transformers-library-has-regular-expression-denial-of)
- [PYSEC-2026-1981 - Hugging Face Transformers is vulnerable to ReDoS through its MarianTokenizer](https://junglewise.ai/threats/cve-2025-6638-hugging-face-transformers-is-vulnerable-to-redos-through-its)