vLLM loads Hugging Face auto_map dynamic modules during model resolution without gating on trust_remote_code, allowing attacker-controlled Python code in a model repo/path to execute at server startup.
An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load.
This happens before any request handling and does not require API access.
All versions where vllm/model_executor/models/registry.py resolves auto_map entries with try_get_class_from_dynamic_module without checking trust_remote_code (at least current main).
During model resolution, vLLM unconditionally iterates auto_map entries from the model config and calls try_get_class_from_dynamic_module, which delegates to Transformers’ get_class_from_dynamic_module and executes the module code.
This occurs even when trust_remote_code is false, allowing a malicious model repo to embed code in a referenced module and have it executed during initialization.
vllm/model_executor/models/registry.py:856 — auto_map resolution vllm/transformers_utils/dynamic_module.py:13 — delegates to get_class_from_dynamic_module, which executes codeReported by bugbunny.ai
| Score | Percentile |
|---|---|
| 0.02% | 6.05% |
| Base score | Version | Severity | Vector |
|---|---|---|---|
| 8.8 | 3.1 | — |
|
| Type | Value |
|---|---|
| GHSA | GHSA-2pc9-4j83-qjmr ↗ |
| CVE | CVE-2026-22807 ↗ |
| CWE id | Name |
|---|---|
| CWE-94 | Improper Control of Generation of Code ('Code Injection') |
Vulnerable version ranges and first patched releases as published by GitHub.
| Ecosystem | Package | Vulnerable range | First patched | Vulnerable functions |
|---|---|---|---|---|
| pip | vllm | >= 0.10.1, < 0.14.0 | 0.14.0 | — |