Vulnerability Details CVE-2026-115912
vLLM is an inference and serving engine for large language models. Prior to 0.32.0, an authenticated caller can place forged shared-memory address and monotonic_id values in public multimodal features.kwargs_data. In scale-out deployments with shared-memory multimodal caching enabled, workers trust those values and pass attacker-selected bytes to pickle.loads(), allowing arbitrary Python execution even when insecure serialization is disabled. This issue is fixed in version 0.32.0.
Exploit prediction scoring system (EPSS) score
EPSS Score 0.005
EPSS Ranking 41.4%
CVSS Severity
CVSS v3 Score 7.5