CVE-2026-54232
HighCVSS 8.8Exploitation Probability (EPSS)
Low risk47th percentile - higher than 47% of all known CVEs
Summary
vLLM, an inference and serving engine for large language models, is vulnerable to a dependency confusion attack in versions prior to 0.22.1. An attacker can register the flashinfer-jit-cache package on PyPI, allowing arbitrary code execution as root during the Docker build.
Risk Assessment
This vulnerability allows an attacker to access sensitive data, such as user prompts, API credentials, and model data, posing a significant threat to the security of production vLLM deployments.
Recommendation
It is recommended to upgrade to vLLM version 0.22.1 or later to mitigate this vulnerability and to block the installation of packages from untrusted sources.
Related vulnerabilities
- CVE-2026-124917Critical
Sandbox escape vulnerability due to use-after-free in the Widget: Gtk component of Firefox. It may allow bypassing process isolation. The vulnerability was fixed in Firefox 160 and Firefox ESR 153.7.
- CVE-2026-124903Critical
A sandbox escape due to a use-after-free exists in the IPC component. This vulnerability was fixed in Firefox 160, Firefox ESR 115.45, and Firefox ESR 153.7.
- CVE-2026-124900Critical
A use-after-free in the DOM: Service Workers component of Firefox allows a sandbox escape. The vulnerability was fixed in Firefox 160.
- CVE-2026-124864Critical
Use-after-free vulnerability in the DOM: Service Workers component of Firefox enabling a sandbox escape. It may allow bypassing browser isolation mechanisms. Fixed in Firefox 160, Firefox ESR 115.45, and Firefox ESR 153.7.
- CVE-2026-124861Critical
Use-after-free vulnerability in the Audio/Video: cubeb component of Firefox enabling a sandbox escape. It may allow bypassing browser isolation mechanisms. Fixed in Firefox 160, Firefox ESR 115.45, and Firefox ESR 153.7.
- CVE-2026-124859Critical
Sandbox escape in the Storage: Cache API component of Firefox caused by incorrect boundary conditions. It may allow bypassing browser isolation mechanisms. Fixed in Firefox 160, Firefox ESR 115.45, and Firefox ESR 153.7.
- CVE-2026-124853Critical
A sandbox escape vulnerability due to incorrect boundary conditions exists in the Graphics: Text component of Firefox. This vulnerability was fixed in Firefox 160, Firefox ESR 115.45, and Firefox ESR 153.7.
- CVE-2026-119446Critical
A Session Fixation vulnerability in the CLI login flow in nerves_hub_web allows an unauthenticated attacker to obtain an API token for another user's account via a crafted login link. The confirmation code is only displayed and never compared, so it cannot prove that the confirming person is the one who started the login.
- CVE-2026-60043Critical
An unauthenticated remote attacker can perform an XSS attack on a logged-in backend user, allowing execution of multiple critical system actions such as shutting down the server or restoring an uploaded backup.
- CVE-2026-13312Critical
The described vulnerability could allow an attacker to compromise a system. Technical details are not provided in the description.
Original NVD description (English source)
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.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS