Vllm CVEs.
Published CVEs affecting Vllm, ranked by exploit odds.
-
CVE-2026-115917
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.
All CVEs affecting Vllm that appear in CVE.org and NVD, sorted by how likely they are to be exploited (EPSS), with fix status and timeline.