← All advisories
CVE-2026-92544highBroken access control2026-11-10

CVE-2026-92544: SQL injection through the model name in Kong AI Gateway and Kong Gateway Enterprise LLM usage counters

Kong Gateway Enterprise and Kong AI Gateway wrote the model name from a client's LLM request into the license_llm_data table without escaping it, so a client could run arbitrary SQL against the Kong database by sending a crafted model name. Kong fixed it in the 10 Nov 2026 Gateway Enterprise patch releases (3.16.0.2, 3.15.0.7, 3.14.0.17, 3.13.0.12, 3.10.0.20) and in Kong AI Gateway 2.3.0 and 2.2.1.

01.

What it is

Kong keeps LLM usage counters for licensing. The model name taken from the client request was put into the SQL statement that writes the license_llm_data table without escaping.

A client that can send requests through an AI route can choose the model name, so it can inject SQL that runs with the gateway's database privileges (CWE-89). The CNA score assumes a low-privileged client and some deployment preconditions (AT:P), with high impact on the Kong database.

Kong's changelogs say the counters now escape every value before running the statement. The fix ships in Kong Gateway Enterprise 3.16.0.2, 3.15.0.7, 3.14.0.17, 3.13.0.12 and 3.10.0.20 (all released 10 Nov 2026) and in Kong AI Gateway 2.3.0 (4 Nov 2026) and 2.2.1 (10 Nov 2026).

02.

How to check

  1. Check your Kong Gateway Enterprise version (kong version, or the Admin API / response header) and compare it with the fixed patch release for your branch.
  2. If you run Kong AI Gateway, check that it is 2.3.0 or later, or 2.2.1 on the 2.2 line.
  3. Review who can send requests to AI proxy routes, since any such client controls the model name.
03.

How to fix

  1. Upgrade to the patched release for your branch: Gateway Enterprise 3.16.0.2, 3.15.0.7, 3.14.0.17, 3.13.0.12 or 3.10.0.20; Kong AI Gateway 2.3.0 or 2.2.1.
  2. Until you can upgrade, limit who can send requests through AI routes to trusted, authenticated consumers.
kong version
# or, from outside: curl -sI https://<your-gateway>/ | grep -i '^server'
04.

Sources

The weekly email

One email a week. Only what changed.

Every new security hole in AI coding tools, MCP servers, app builders and the stacks they generate (Next.js, Supabase, Node, PHP), each with how to check your app and the fix. No roundups of things you already read.

Free, one email a week. Read the latest issue first if you like. One-click unsubscribe in every email. We never sell or share addresses.