Hugging Face Transformers Vulnerability Exposes Users to Remote Code Execution

A vulnerability in the Hugging Face Transformers library (versions 4.49.0 to 5.8.1) allows remote attacker-controlled Python files to be written to users' local disks without consent. This issue, tracked as CVE-2026-80047, arises from the library's custom generation-loading process, which fetches and caches remote modules before evaluating user consent. Although the execution of the module is gated, the initial file write occurs unconditionally, leading to unauthorized files being stored in the user's cache. Attackers can exploit this flaw by publishing malicious model repositories, which can result in persistent unauthorized code on user systems. Currently, no patch or advisory has been issued by the vendor, and users are advised to avoid untrusted model repositories and regularly inspect their cache. The vulnerability poses a significant risk as it can lead to unintended execution of cached code during trusted model loads.

Key Points: • Vulnerability allows unauthorized Python files to be written to disk without user consent. • Affected versions of Hugging Face Transformers are 4.49.0 through 5.8.1. • No patch or vendor advisory is currently available; users should inspect their cache.