vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the OpenAI-compatible server configures Starlette CORSMiddleware with allowed_origins set to a wildcard by default. A malicious web page can use a victims browser to reach an unauthenticated vLLM server on localhost, an internal network, or a VPN, call endpoints including GET /v1/models and POST /v1/chat/completions, consume GPU resources, and read responses. Cookie-based cross-site request forgery does not apply because credentials are disabled by default, and deployments requiring bearer authentication are exposed only when the caller already has a usable token. This issue is fixed in version 0.32.0.
Weakness
The product uses a web-client protection mechanism such as a Content Security Policy (CSP) or cross-domain policy file, but the policy includes untrusted domains with which the web client is allowed to communicate.
Affected Software
| Name | Vendor | Start Version | End Version |
|---|---|---|---|
| Vllm | Vllm | * | 0.32.0 (excluding) |
Extended Description
If a cross-domain policy file includes domains that should not be trusted, such as when using wildcards under a high-level domain, then the application could be attacked by these untrusted domains. In many cases, the attack can be launched without the victim even being aware of it.