CVE-2026-105755
Medium · 4.2CVSS
4.2
EPSS
0.2%
Published
Oct 5, 2026
CISA due date
—
The record
vLLM is an inference and serving engine for large language models. Prior to 0.30.0, flash late-interaction scoring at the /score and /rerank endpoints derives each worker's query_key value from the caller-controlled X-Request-Id header. A concurrent request that reuses a victim's identifier can overwrite the cached query embedding so the victim's documents are scored against the attacker's query, and shared use counters can also cause a late-interaction cache-miss error. This issue is fixed in version 0.30.0.
Affected packages
| Ecosystem | Package | Introduced | Fixed |
|---|
Related vulnerabilities
- CVE-2026-13319 GitLab has remediated an issue in GitLab CE/EE affecting all versions from 15.10 before… CWE-639
- CVE-2026-111767 MCP TypeScript SDK is the official TypeScript SDK for Model Context Protocol servers and… CWE-639
- CVE-2026-105754 vLLM is an inference and serving engine for large language models. CWE-639
- CVE-2026-62140 Unauthenticated Insecure Direct Object References (IDOR) in Quiz And Survey Master <=… CWE-639
- CVE-2026-62134 Contributor Insecure Direct Object References (IDOR) in Starter Templates <= 4.7.5… CWE-639
- CVE-2026-62113 Contributor Insecure Direct Object References (IDOR) in Slim SEO <= 4.10.0 versions. CWE-639
- CVE-2026-89264 MoguBlog through 6.2 fails to validate the comment author identity in the POST… CWE-639
- CVE-2026-89262 MoguBlog through 6.2 contains an authorization bypass vulnerability in the comment… CWE-639