vLLM P2P KV offloading denial of service vulnerability

# vLLM P2P KV offloading denial of service vulnerability

https://gcdradar.vercel.app/incidents/vllm-p2p-kv-offloading-denial-of-service-vulnerability · AIR-2026-0155 · closed 2026-09-21

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Patched
- First reported: 2026-09-21
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2026-94624
- CVSS: 8.7

## Summary

A denial of service vulnerability in vLLM P2P KV offloading allows supplying arbitrary remote host and port values to exhaust ZeroMQ sockets. This triggers an uncaught ZMQError that crashes EngineCore and stops all inference.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Affected and fixed versions

- Affected: vllm < 0.32.0
- Fixed in: vllm 0.32.0

## Latest updates

- 2026-10-27: GitHub advisory published (https://github.com/advisories/GHSA-frvg-qw88-hww9)
- 2026-09-22: GitHub advisory published (https://github.com/advisories/GHSA-8qpc-6rxx-4pxf)
- 2026-09-21: CVE-2026-94624: vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with… (https://nvd.nist.gov/vuln/detail/CVE-2026-94624)

## Sources

- 2026-09-21 NVD: CVE-2026-94624: vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with… https://nvd.nist.gov/vuln/detail/CVE-2026-94624
- 2026-09-22 GitHub Advisory Database: GHSA-8qpc-6rxx-4pxf: Duplicate Advisory: vllm p2p never connected zmq_exhaustion https://github.com/advisories/GHSA-8qpc-6rxx-4pxf
- 2026-10-27 GitHub Advisory Database: CVE-2026-94624: vllm p2p never connected zmq_exhaustion https://github.com/advisories/GHSA-frvg-qw88-hww9

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.