Zero Radar
← feed
HighVulnerabilityupdated 30 Oct 2026, 14:17 UTC

vLLM: several flaws let attackers leak data, tamper, or crash the service

vLLM is software for running AI language models on a server. Several weaknesses could let an attacker read information they should not see, change data, bypass security checks, or make the service stop working.

Official source
SeverityHighVulnerability
Being exploited?UnclearNo reports either way yet
Fix available?UnclearNo fixed version published in the sources yet
How sure is this?Official sourceBacked by an official source

What happened

Multiple vulnerabilities in vLLM allow an attacker to disclose information, manipulate data, bypass security measures, or cause a denial-of-service condition. The advisory does not specify the individual flaws.

Why it is dangerous: Successful exploitation could expose sensitive data handled by the AI service, alter its results, or take it offline. Organizations running vLLM as an AI inference server are at risk.

Are you affected?

You are affected if you run vLLM (affected versions not stated in the material).

vllm (specific versions unknown)

What to do

  1. Check whether vLLM is deployed in your environment and which version is running
  2. Consult the official vLLM release notes or the CERT-Bund advisory for a fixed version and update when available
  3. Restrict network access to the vLLM service until it is patched

Timeline

  1. 30 Oct 2026, 14:17 UTC
    Publicly disclosed
  2. 30 Oct 2026, 14:17 UTC
    Reported by CERT-Bund (BSI) · CERT-Bund (BSI)
  3. 30 Oct 2026, 14:17 UTC
    Reported by CERT-Bund (BSI) · CERT-Bund (BSI)
  4. 30 Oct 2026, 14:17 UTC
    Picked up by Zero Radar

Sources 2 reports, oldest first

Official vendor, government CERT or advisory database

Technical details

ID
ZR-edce1356 · ZR-fbeba36e
Disclosed
30 Oct 2026, 14:17 UTC
On Zero Radar
30 Oct 2026, 14:17 UTC (1 min after disclosure)
Last change
30 Oct 2026, 14:17 UTC

The plain-language parts and the attack diagram were written by an AI model (anthropic/claude-haiku-5.5) from the sources above. Facts like versions, exploitation and dates come from the sources themselves. How Zero Radar works

vLLM: several flaws let attackers leak data, tamper, or crash the service · Zero Radar