CVE-2026-115917: vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the OpenAI-compa…
- Occurred
- Oct 26, 2026
- Published
- Oct 26, 2026
- Source
- NVD CVE
Summary
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 victim's 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.
Tags
- CVE
- CVE-2026-115917
- cvss:LOW
Classification
Rationale: Unauthenticated access to LLM endpoints via CORS misconfiguration allows unauthorized resource consumption and data exposure.
Classifier: Qwen/Qwen3.5-35B-A3B-FP8