Hugging Face Transformers Flaw Writes Malicious Python Code to Disk Before User Consent

A newly disclosed vulnerability (CVE-2026-80047) in Hugging Face Transformers (affecting versions 4.49.0 through 5.8.1) causes the library to download and write remote custom-generation Python modules to the local cache before confirming user consent via trust_remote_code; this allows attacker-controlled files to be persisted and potentially lead to later code execution in shared or CI/CD environments. CERT/CC highlighted the issue stems from an unconditional file-copy operation in dynamic_module_utils.py, noted the absence of a vendor patch at publication, and recommended avoiding load_custom_generate() with untrusted repositories, inspecting and clearing the Hugging Face module cache, and ensuring trust checks occur prior to writing untrusted code.