AI Stack Alert High CVE-2026-116180 For AI Eng, AppSec, CISO
CVE-2026-116180: vLLM from 0.8.5 before 0.31.0 has a denial of service flaw. LogprobsTensors.empty_cpu() in vllm/v1/outp
CVE-2026-116180 vLLM from 0.8.5 before 0.31.0 has a denial of service flaw. LogprobsTensors.empty_cpu() in vllm/v1/outputs.py allocates an uninitialized int32 tensor, and the V1 model runner leaves prompt positions reported as computed by a KV connector unwritten.
Why it matters
AI tooling often runs with broad access to code, data and credentials, which raises the impact of this issue.
Do this
- Check whether the affected component is used in your environment and track the vendor fix.
- Review exposure: restrict network access and permissions of the affected component until patched.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.25%
Signals
- NoCISA KEV
- 8.3CVSS
- 0.25%EPSS
- NoPublic PoC
- 35/100Priority
- 75%Confidence
Evidence
“CVE-2026-116180 vLLM from 0.8.5 before 0.31.0 has a denial of service flaw. LogprobsTensors.empty_cpu() in vllm/v1/outputs.py allocates an uninitialized int32 tensor, and the V1 model runner leaves prompt positions reported as computed by a KV connector unwritten.”
Sources
Affected / related
vLLM, vllm
Published 2026-10-25 14:16 UTC · updated 2026-11-08 02:48 UTC · How we score