CVE-2026-96437

Large prompts could bypass configured content-safety inspection

Kong · Gateway Enterprise Azure AI content-safety plugin

First documented disclosure · Review checkpoint

Summary

The safety plugin skipped request bodies that exceeded its in-memory buffer. A request could reach the upstream model without the configured input inspection.

Severity and exploitation evidence

High · 8.2/10 · CVSS 4.0. Severity source.

Exploitation not established: The reviewed advisory does not report malicious exploitation.

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 safety plugin skipped request bodies that exceeded its in-memory buffer.

Conditions required

A larger prompt reached a deployment using the Azure AI content-safety plugin.

Potential impact

A request could reach the upstream model without the configured input inspection.

Affected versions and configurations

Affected release lines begin at 3.13.0.0, 3.14.0.0, 3.15.0.0 and 3.16.0.0 and end before the listed fixes.

Fix and mitigation status

3.13.0.12, 3.14.0.11, 3.15.0.4 and 3.16.0.1 respectively.

Practical next steps

Apply the fix for the deployed branch and verify handling of oversized requests.

Operating lesson

Verify content inspection for both buffered and larger request bodies.

Evidence and limits

The CVE record was published on 2026-11-10.

Sources