VDB

CVE-2026-105755

CVE-2026-105755 PUBLISHED CVSS 4.2 MEDIUM

Reported by GitHub_M · Published October 5, 2026

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.

Risk Scores

CVSS 3.1
4.2
CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:L/A:L

Affected Products

VendorProductVersions
vllm-projectvllm< 030.0
PyPIvllm0
chainguardtritonserver-backend-vllm-cuda-13.00
vllm-projectvllm0
chainguardvllm-openai-cuda-13.00
chainguardvllm-cuda-13.20

Timeline

  • Oct 5, 2026 CVE Published
  • Oct 6, 2026 EPSS Score
  • Oct 6, 2026 Coalition ESS Score
  • Oct 6, 2026 CVE Updated
  • Oct 8, 2026 EPSS Score
  • Oct 13, 2026 EPSS Score
  • Oct 19, 2026 EPSS Score
  • Oct 25, 2026 EPSS Score
  • Oct 29, 2026 EPSS Score
  • Oct 31, 2026 Security Advisory
Open in Interactive Console →
$ Console Community · 50/day Open console ›