Use of Less Trusted Source Affecting vllm package, versions [0.22.1,0.28.0)
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 applications- Snyk IDSNYK-PYTHON-VLLM-20158365
- published27 Sept 2026
- disclosed26 Sept 2026
- creditrexpository
Introduced: 26 Sep 2026
NewCVE-2026-100653 (opens in a new tab)How to fix?
Upgrade vllm to version 0.28.0 or higher.
Overview
vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs
Affected versions of this package are vulnerable to Use of Less Trusted Source via incomplete propagation of revision and code_revision pins across artifact boundaries in the model loader subsystem. When an operator supplies a revision pin to lock a model to a specific version, several artifact-loading paths - including gguf_loader.py, kimi_audio.py, kimi_k25.py, registry.py, and roberta.py - ignore that pin and resolve config files, weight files, image processors, and sub-model artifacts against the default (latest) revision instead. An attacker who can influence the default-revision content of a Hugging Face repository can cause a pinned deployment to silently load unpinned, potentially malicious artifacts, achieving partial confidentiality impact and high integrity impact.
Note: This is only exploitable when an operator supplies a revision or code_revision pin and the attacker can influence the content served at the unpinned (default) revision of the target repository.