A critical remote code execution vulnerability, CVE-2026-105192 with a CVSS score of 9.8, has been identified in LMCache, a key-value cache server utilized by vLLM. The flaw, present in versions 0.3.9 through 0.5.5, allows unauthenticated attackers to execute arbitrary code as root by exploiting an exposed ZeroMQ socket in multiprocess mode. As of the reports, no patch is available for this vulnerability. AI
IMPACT This critical vulnerability in LMCache, used by vLLM, poses a significant security risk to AI deployments, potentially allowing unauthorized root access.
RANK_REASON Security vulnerability disclosure in an AI infrastructure component.
Read on Mastodon — mastodon.social →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →