CVE-2026-126397

Cohere2 Vision image processing could monopolize shared CPU workers

vLLM · Cohere2 Vision processor

First documented disclosure · Review checkpoint

Summary

The service passed a client-selected patch count to an image processor whose work grows quadratically. Image preprocessing could deny capacity to other clients before model inference even begins.

Severity and exploitation evidence

High · 8.7/10 · CVSS 4.0. Severity source.

Exploitation not established: The reviewed sources describe the weakness and correction. They identify no confirmed malicious use.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

The service passed a client-selected patch count to an image processor whose work grows quadratically.

Conditions required

A client can call an affected vision deployment with request-level processing options.

Potential impact

Image preprocessing could deny capacity to other clients before model inference even begins.

Affected versions and configurations

0.10.1 through releases before 0.31.0.

Fix and mitigation status

0.31.0 for this issue.

Practical next steps

Use a supported vLLM release containing the 0.31.0 fix. Keep client-controlled multimodal settings disabled for untrusted users.

Operating lesson

Apply resource limits to preprocessing workers as well as inference workers.

Evidence and limits

A September mitigation covered a class of multimodal settings. The October project notice describes this processor-specific exposure. The issuing CNA published the CVE record on 2026-11-13.

Sources