Medium [CVE-2026-54232] Arbitrary code execution via dependency confusion during Docker build

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.1, the vLLM Dockerfile is vulnerable to a dependency confusion attack through the flashinfer-jit-cache package.

The package is installed from a custom index (flashinfer.ai/whl/) using --extra-index-url, but the package name was not registered on PyPI, and UV_INDEX_STRATEGY="unsafe-best-match" is set globally.

An attacker who registers flashinfer-jit-cache on PyPI with version 0.6.11.post2 can execute arbitrary code as root during the Docker build and backdoor every resulting container image, enabling exfiltration of all user prompts, API credentials, and model data from production vLLM deployments This vulnerability is fixed in 0.22.1.

CVE-2026-54232 is a build-time dependency confusion issue in upstream vLLM Dockerfiles before 0.22.1. It does not allow remote exploitation of a running vLLM inference service.

Red Hat OpenShift AI is not affected. Red Hat AI Inference Server and RHEL AI CUDA images that include flashinfer-jit-cache are in scope for build-process review, but Red Hat has no evidence that shipped images were compromised.

Red Hat severity: Moderate — CVSS 5.7 (CVSS:3.1/AV:N/AC:H/PR:H/UI:R/S:U/C:H/I:H/A:N). Weakness: CWE-426.

Affected Red Hat products: Red Hat AI Inference Server; Red Hat Enterprise Linux AI (RHEL AI) 3.