CVE-2026-105754

/// Technical 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.

/// Remediation

Check the references section for vendor advisories and patches from vllm. Update vllm to the latest patched version. If immediate patching is not possible, review the CVSS vector to understand the attack surface and apply compensating controls such as network segmentation or access restrictions.

/// Affected products

/// References