ZDF
Sign inSign up

Vulnerabilities

Vulnerability · CVE-2026-100653

vLLM is an inference and serving engine for large language models.

Published Sep 26, 2026 · catalog updated Oct 20, 2026 · vLLM is an inference

Open the advisory

Exploitation likelihood is not on this record for CVE-2026-100653. A missing EPSS score is not a low risk. Compare vLLM is an inference with the linked advisory before you change production.

Gaps on this record

  • The stored product name is a package slug. The name on this page is taken from the title (vLLM is an inference).
  • No vendor fix or required action is stored. Compare the linked advisory before you change production.
  • EPSS is missing. A missing score is not a low risk.
Severity
High8.3
Known exploitation
Not in CISA/VulnCheck KEV
EPSS
Awaiting synchronization with FIRST
Remediation
Version ranges on file
Affected products
cpe vllm/vllm

vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for the FunAudioChat and Tarsier2 architectures: the WhisperFeatureExtractor and speech_tokenizer PreTrainedTokenizerFast loads in vllm/model_executor/models/funaudiochat.py and the Qwen2VLConfig.from_pretrained call used by Tarsier2ProcessingInfo in vllm/model_executor/models/qwen2_vl.py. As a result, deployments pinned to a reviewed revision still resolve these behavior-affecting processor, tokenizer, and config artifacts from the repository's default revision, so a later change to the upstream default branch can alter audio preprocessing, speech tokenizer behavior, or Tarsier2 configuration without any change to the operator's configured pin. This is a supply-chain integrity and reproducibility failure for pinned deployments; it is residual to the earlier fix tracked as GHSA-3ww4-5jv9-j5gm / CVE-2026-47155 and does not constitute remote code execution or a trust_remote_code=False bypass. The issue is fixed in version 0.28.0.

Are you affected?

Enter the version you run. The result uses only the ranges stored on this record.

The check returns affected, not affected, or incomplete. It does not guess a version that is missing from the record.

  • cpe vllm/vllm< 0.28.0

How to fix CVE-2026-100653

What to do

Three questions, each answered from this record only. Confirm the vendor advisory before you patch.

Should you patch?

Exploitation likelihood is not on this record for CVE-2026-100653. A missing EPSS score is not a low risk. Compare vLLM is an inference with the linked advisory before you change production.

Which versions are affected?

Compare what you have installed with < 0.28.0. Compare the installed version with the linked vendor advisory before you change production.

What should an analyst verify?

CWE-348. EPSS is not on this record yet. There is no CISA KEV listing. No Exploit-DB or Metasploit reference is linked. Use the source links for the original advisory.

Code

The project this record applies to. Exploit listings stay in References.

Timeline

Detection, updates, and resolution for this record.

  1. Sep 26, 2026 · NIST NVD

    Detected

    vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for the FunAudioChat and Tarsier2 architectures: the WhisperFeatureExtractor and speech_tokenizer PreTrainedTokenizerFast loads in vllm/model_executor/models/funaud

  2. Oct 20, 2026 · NIST NVD

    Updated

    vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for the FunAudioChat and Tarsier2 architectures: the WhisperFeatureExtractor and speech_tokenizer PreTrainedTokenizerFast loads in vllm/model_executor/models/funaud

Sources

  • CVE.orgVulnerability recordCVE-2026-100653

    Official CVE identifier

    cve.org
  • NIST NVDVulnerability recordCVE-2026-100653

    vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for the FunAudioChat and Tarsier2 architectu

    nvd.nist.gov

References

Freshness

Last verified in this catalog Oct 20, 2026. Significant claims (KEV, EPSS, CVSS, affected ranges) come from the source rows above — open the original publisher for the authoritative advisory.