vllm: Schwachstelle ermöglicht Codeausführung
WID-SEC-2026-0890 · first 27.03.2026 · updated 10.07.2026 · CVSS 8,8
What it is about
Open Source vLLM ist eine Open-Source-Bibliothek für schnelle und effiziente Inferenz von Large Language Models (LLMs). Ein entfernter, anonymer Angreifer kann eine Schwachstelle in vllm ausnutzen, um beliebigen Programmcode auszuführen.
Under what conditions?
Over the network and without logging in. This is the most dangerous combination: attackers can scan reachable devices automatically.
What can happen in the worst case
An attacker could use the vulnerability to
- beliebigen Programmcode auszuführen
Am I affected?
Affected according to the BSI:
- Red Hat Enterprise Linux
- Open Source vllm <0.18.0
The installed version is usually shown under “Help → About” or in the device’s admin interface.
What helps?
Update to at least:
- Open Source vllm 0.18.0
Is the vulnerability already being exploited?
None of the CVEs is in the CISA catalogue of exploited vulnerabilities. That does not mean there are no attacks, only that none are documented.
Information and sources
- Original BSI advisory (WID)
- vLLM GitHub vom 2026-03-26 github.com
- Red Hat Security Advisory RHSA-2026:10141 vom 2026-04-23 access.redhat.com
- Red Hat Security Advisory RHSA-2026:10140 vom 2026-04-23 access.redhat.com
- Red Hat Security Advisory RHSA-2026:19724 vom 2026-05-21 access.redhat.com
- Red Hat Security Advisory RHSA-2026:19712 vom 2026-05-21 access.redhat.com
- Red Hat Security Advisory RHSA-2026:24977 vom 2026-06-10 access.redhat.com
- Red Hat Security Advisory RHSA-2026:37275 vom 2026-07-09 access.redhat.com
CVE: CVE-2026-27893
Source: BSI warning and information service (CERT-Bund); CISA. The assessment of the conditions is my reading of the BSI text; the external pages are in their original language.