vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features

[ Description ]Official 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.

[ Fixes ]Fixes and patches

• vllm 0.30.0Fixed version · GitHub Advisories · pip package vllm: upgrade to 0.30.0 or later (affected < 0.30.0)
• Patch on github.comVendor patch or bulletin · NVD · Reference tagged 'Patch' by NVD
• Patch on github.comVendor patch or bulletin · NVD · Reference tagged 'Patch' by NVD

[ Advisories ]Government CERT advisories

• BSI CERT-Bund23 Sep 2026WID-SEC-2026-3564

  vllm: Multiple vulnerabilities Translated from German

  CVE-2026-105757CVE-2026-105754CVE-2026-105755+1 more CVEs

[ Packages ]Affected open-source packages

• vllm pipAffected < 0.30.0 · fixed in 0.30.0 · GHSA-ph72-cqr5-qpp7

[ References ]References

• github.com/vllm-project/vllm/commit/1970f3ed4be7fa8620e4ddc4a12c36a8384cfc27Patch
• github.com/vllm-project/vllm/pull/51898Issue TrackingPatch
• github.com/vllm-project/vllm/releases/tag/v0.30.0Release Notes
• github.com/vllm-project/vllm/security/advisories/GHSA-ph72-cqr5-qpp7MitigationVendor Advisory