# CVE-2026-57845: Hugging Face Datasets insecure deserialization in pandas builder

Source: https://junglewise.ai/threats/cve-2026-57845-hugging-face-datasets-through-5-1-0-contains-an-insecure  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-23. Updated: 2026-10-23  
License: CC BY 4.0 with a link back

CVE-2026-57845 · Severity: high · CVSS 8.8 · Published 2026-10-23

Hugging Face Datasets is a library for loading and processing machine learning datasets. Versions through 5.1.0 contain a critical flaw in the pandas dataset loader that allows attackers to execute arbitrary Python code when a user loads a malicious dataset file. An attacker can craft a poisoned pickle file that runs code with the victim's privileges, potentially stealing credentials, modifying data, or compromising the entire system.

Technologies: [Hugging Face Datasets](https://junglewise.ai/threats/technologies/datasets). Vendors: [Hugging Face](https://junglewise.ai/threats/vendors/hugging-face).

## Quick answers

**Is CVE-2026-57845 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-57845 being exploited in the wild.

**How severe is CVE-2026-57845?** High severity, with a CVSS score of 8.8, published 23 October 2026.

**Which products and versions are affected?** Hugging Face Datasets through 5.1.0.

**Is there a fix for CVE-2026-57845?** No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.

**How is CVE-2026-57845 attacked?** It is attacked over the network, in a software library.

**Am I affected by CVE-2026-57845?** Check whether you run Hugging Face Datasets 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.

## Technical details

The pandas packaged builder in Hugging Face Datasets passes user-supplied pickle files directly to pandas.read_pickle() without validation, triggering deserialization of untrusted data. The vulnerability is in datasets.packaged_modules.pandas.Pandas._generate_tables(), which opens attacker-controlled files and unpickles them, executing arbitrary Python code embedded in the pickle's __reduce__ method before any data validation occurs. Exploitation requires user interaction to call datasets.load_dataset("pandas", data_files=...) with attacker-supplied paths, but no authentication or network privilege is required.

## Affected products

- Hugging Face Datasets through 5.1.0

## Timeline

- 2026-10-23: disclosed

## References

- [https://github.com/SaiTeja-Erukude/CVE-2026-57845-hf-datasets-rce](https://github.com/SaiTeja-Erukude/CVE-2026-57845-hf-datasets-rce)
- [https://github.com/huggingface/datasets/pull/8879](https://github.com/huggingface/datasets/pull/8879)

## Related threats

- [Hugging Face Datasets path traversal in folder-based builders](https://junglewise.ai/threats/cve-2026-66007-hugging-face-datasets-path-traversal-in-folder-based-builders)
- [Hugging Face Datasets symlink following in Extractor.extract](https://junglewise.ai/threats/cve-2026-65010-hugging-face-datasets-symlink-following-in-extractor-extract)