Incorrect Authorization Affecting transformers package, versions [4.49.0,]
Threat Intelligence
The probability is the direct output of the EPSS model, and conveys an overall sense of the threat of exploitation in the wild. The percentile measures the EPSS probability relative to all known EPSS scores. Note: This data is updated daily, relying on the latest available EPSS model version. Check out the EPSS documentation for more details.
Do your applications use this vulnerable package?
In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.
Test your applicationsSnyk Learn
Learn about Incorrect Authorization vulnerabilities in an interactive lesson.
Start learning- Snyk IDSNYK-PYTHON-TRANSFORMERS-19500670
- published2 Sept 2026
- disclosed1 Sept 2026
- creditPrasanna Dabi
Introduced: 1 Sep 2026
NewCVE-2026-80047 (opens in a new tab)How to fix?
There is no fixed version for transformers.
Overview
transformers is a State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow
Affected versions of this package are vulnerable to Incorrect Authorization in the load_custom_generate() method of GenerativePreTrainedModel, where get_cached_module_file() in dynamic_module_utils.py copies the model's custom_generate/generate.py into the local module cache before the resolve_trust_remote_code() consent check runs. An attacker can plant a Python file in the victim's cache at ~/.cache/huggingface/modules even after the user declines the trust prompt, priming it for execution during a later trusted operation that reuses the cache, by publishing a model repository containing a malicious custom_generate/generate.py. Only the file write bypasses authorization, since execution of the cached module remains gated by the consent check, and exploitation requires the victim to call load_custom_generate() on the attacker's repository.
Workaround
This vulnerability can be avoided by regularly inspecting and clearing the Hugging Face module cache at ~/.cache/huggingface/modules, which removes any file written before the trust prompt was declined.