CVE-2026-100653
MEDIUMDescription
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.
How to fix
TridentStack Control can deploy fixes like this automatically across your Windows, macOS, and Linux fleet. See how it works
Remediation is compiled from vendor and distribution security advisories. Always confirm against the linked source for your exact version and platform.
CVSS v3 Vector
Exploitability
Impact
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:H/A:N
Exploit Intelligence
Low risk: more likely to be exploited than 27% of all known CVEs.
References
Related Vulnerabilities
Other CWE-348 vulnerabilities, ordered by exploit likelihood. View all
| CVE | Severity | CVSS | EPSS | Exploited | Fix |
|---|---|---|---|---|---|
| CVE-2025-59951 | Critical | 9.1 | 4.5% | - | - |
| CVE-2022-31813 | Critical | 9.8 | 3.5% | - | Fix |
| CVE-2024-45410 | Critical | 9.8 | 1.5% | - | Fix |
| CVE-2021-21373 | High | 7.5 | 1.2% | - | Fix |
| CVE-2021-21374 | High | 8.1 | 1.0% | - | Fix |
| CVE-2022-2255 | High | 7.5 | 0.9% | - | Fix |
Common questions
How do I fix CVE-2026-100653?
Upgrade vllm to 0.28.0 or later.
Is CVE-2026-100653 being actively exploited?
Not that we know of. CVE-2026-100653 is not in the CISA Known Exploited Vulnerabilities catalog. Its EPSS score of 0.34% is the estimated probability that it will be exploited in the next 30 days. That is higher than 27% of all scored CVEs.
How severe is CVE-2026-100653?
CVE-2026-100653 has a CVSS v3 base score of 6.5, rated medium. CVSS rates the technical impact if the vulnerability is exploited, not how likely that is, so weigh it alongside the exploit-prediction score when you decide what to patch first.
What does CVE-2026-100653 affect?
Published advisories record a fix for vllm (NVD). Only products with a sourced advisory are listed, so treat this as what we can cite rather than a complete inventory.
Embed a live status badge for CVE-2026-100653
Markdown
[](https://tridentstack.com/cve/CVE-2026-100653)HTML
<a href="https://tridentstack.com/cve/CVE-2026-100653"><img src="https://tridentstack.com/cve/badge/CVE-2026-100653.svg" alt="CVE-2026-100653"></a>Find and fix vulnerabilities across your fleet
TridentStack Control continuously scans your Windows, macOS, and Linux fleet for known vulnerabilities, prioritizes them by severity and active exploitation, and patches them automatically.
This product uses NVD data but is not endorsed or certified by the NVD. EPSS scores courtesy of FIRST.org (https://www.first.org/epss). Source: CISA KEV Catalog. Data as of 2026-10-20.