CVE-2026-100653

Published 26 September 2026 · tracked since 27 September 2026

Description

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 WhisperFeat

References

• NVD — National Vulnerability Database
• CVE.org record (MITRE)
• CISA Known Exploited Vulnerabilities catalog

Latest tracked vulnerabilities

• CVE-2026-100876 MEDIUM 6.3
• CVE-2026-100875 HIGH 7.3
• CVE-2026-100873 MEDIUM 4.3
• CVE-2026-100874 HIGH 7.3
• CVE-2026-101060 HIGH 8.2
• CVE-2026-101061 MEDIUM 4.7
• CVE-2026-101047 MEDIUM 5.3
• CVE-2026-101057 LOW 3.1
• CVE-2026-101058 MEDIUM 6.9
• CVE-2026-101044 HIGH 7.1

→ Open the live CVE board · all tools