AI Stack Alert Low CVE-2026-129356 For AI Eng, AppSec, SOC / IR AI-assisted
vLLM 0.30.0/0.31.0 race condition in single_type_kv_cache_manager.py (CVE-2026-129356)
A race condition was found in vllm-project vllm versions 0.30.0 and 0.31.0, in the Common Prefix Handler component, file single_type_kv_cache_manager.py. The source states the attack may be performed remotely but is highly complex and difficult to exploit, and that a public exploit exists. Upgrading to version 0.32.0 resolves the issue, and the patch is identified as e687a251bb056fa527846633f7a1b0a6c0f42374.
Why it matters
Organizations running vLLM 0.30.0 or 0.31.0 to serve models are exposed to this flaw, which carries a low CVSS base score of 1.3. Because a public exploit exists, unpatched inference servers should be prioritized for upgrade.
Do this
- Inventory deployments of vLLM and identify any running version 0.30.0 or 0.31.0.
- Upgrade affected vLLM instances to version 0.32.0 or later.
- Verify the fix by confirming commit e687a251bb056fa527846633f7a1b0a6c0f42374 is included in the deployed build.
- Restrict network exposure of vLLM inference endpoints until patched.
Exploit Clock
No signals yet
- No exploitation signals yet
Signals
- NoCISA KEV
- 1.3CVSS
- NoPublic PoC
- 18/100Priority
- 60%Confidence
Evidence
“Upgrading to version 0.32.0 is able to resolve this issue. The patch is identified as e687a251bb056fa527846633f7a1b0a6c0f42374.”
“The exploit has been made public and could be used.”
“It is stated that the exploitability is difficult.”
Sources
Affected / related
vllm, race condition
Published 2026-11-15 02:17 UTC · updated 2026-11-15 06:35 UTC · AI-assisted summary, checked against quoted sources · How we score