CVE-2026-69147: vLLM video backend GPU memory allocation flaw

# CVE-2026-69147: vLLM video backend GPU memory allocation flaw

https://gcdradar.vercel.app/incidents/cve-2026-69147-vllm-video-backend-gpu-memory-allocation-flaw · AIR-2026-0138 · closed 2026-09-16

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2026-09-16
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2026-69147
- CVSS: 6.5

## Summary

A vulnerability in vLLM prior to version 0.28.0 allows chat completion requests to force unexpected video backend decoders. This bypasses static GPU memory budgeting and can lead to unmanaged CUDA allocations.

## What to do

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

## Latest updates

- 2026-09-17: CVE-2026-69147: vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservation (https://github.com/advisories/GHSA-8pw2-6jv3-mj5j)
- 2026-09-16: CVE-2026-69147: vLLM is an inference and serving engine for large language models. (https://nvd.nist.gov/vuln/detail/CVE-2026-69147)

## Sources

- 2026-09-16 NVD: CVE-2026-69147: vLLM is an inference and serving engine for large language models. https://nvd.nist.gov/vuln/detail/CVE-2026-69147
- 2026-09-17 GitHub Advisory Database: CVE-2026-69147: vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservation https://github.com/advisories/GHSA-8pw2-6jv3-mj5j

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