AI Stack Alert High CVE-2026-126394 For AI Eng, SOC / IR, AppSec, CISO AI-assisted
CVE-2026-126394: vLLM 0.30.0 to before 0.31.0 resource exhaustion via request-controlled num_frames
vLLM versions 0.30.0 before 0.31.0 contain a resource exhaustion flaw in OpenCVDynamicOpenPanguVideoBackend. The backend uses a request-controlled num_frames value without an effective ceiling. Unauthenticated requests with a large num_frames can force large allocations and CPU work that OOM-kill the API server.
Why it matters
Any exposed vLLM API server running the affected versions can be crashed by unauthenticated requests, disrupting model serving for dependent applications. The CVSS base score is 6.9.
Do this
- Upgrade vLLM to version 0.31.0 or later.
- Until patched, restrict access to /v1/chat/completions and /tokenize to trusted clients at the network or gateway layer.
- Monitor API server memory and OOM-kill events and alert on unexpected restarts.
- Review whether the openpangu video_backend is enabled and disable video processing where it is not needed.
Exploit Clock
No signals yet
- No exploitation signals yet
Signals
- NoCISA KEV
- 6.9CVSS
- NoPublic PoC
- 36/100Priority
- 80%Confidence
Evidence
“vLLM 0.30.0 before 0.31.0 contains a resource exhaustion vulnerability in OpenCVDynamicOpenPanguVideoBackend”
“Unauthenticated attackers can send /v1/chat/completions or /tokenize requests selecting the openpangu video_backend with a huge num_frames”
“forcing np.linspace allocations and CPU work that OOM-kill the API server.”
Sources
Affected / related
vLLM, resource exhaustion
Published 2026-11-13 10:17 UTC · updated 2026-11-13 12:35 UTC · AI-assisted summary, checked against quoted sources · How we score