Vllm-project - Vllm - CRITICAL - CVE-2026-48746. A vulnerability exists in the vLLM inference engine affecting OpenAI API integration, enabling an authentication bypass due to misconfigurations in ASGI web servers and the Trust feature in Starlette. This flaw permits unauthorized access to the API, allowing interaction without the necessity for valid credentials such as VLLM_API_KEY. The issue has been addressed and mitigated in version 0.22.0, emphasizing the need for users to update their installations in order to secure their API interactions.