VYPR
Medium severity6.5NVD Advisory· Published Oct 5, 2026· Updated Oct 8, 2026

CVE-2026-105754

CVE-2026-105754

Description

vLLM is an inference and serving engine for large language models. Prior to 0.30.0, the /inference/v1/generate endpoint in the disaggregated scale-out path accepts caller-supplied tensors in the features.kwargs_data field, cache identifiers in the features.mm_hashes field, ranges in the features.mm_placeholders field, and wire-selected multimodal field processors without rebinding them to the active model renderer contract. Forged grid geometry, field types, or non-positive placeholder lengths can terminate the shared EngineCore; when an attacker knows or can induce a victim's content hash, forged cache hashes can poison or retrieve cross-request encoder-cache state; and dropped sparse placeholder masks can alter replayed transport semantics. This issue is fixed in version 0.30.0.

AI Insight

LLM-synthesized narrative grounded in this CVE's description and references.

Affected packages

Versions sourced from the GitHub Security Advisory.

PackageAffected versionsPatched versions
vllmPyPI
< 0.30.00.30.0

Affected products

2
  • Vllm/Vllm2 versions
    cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*+ 1 more
    • cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*range: <0.30.0
    • (no CPE)range: <0.30.0

Patches

Vulnerability mechanics

References

5

News mentions

1