CVE-2026-53998

Phoenix chat templates could expose secrets and enable administrator session forgery

Arize-ai · Arize Phoenix

First documented disclosure · Review checkpoint

Summary

Chat-template evaluation could reveal process secrets, including the key protecting administrator sessions.

Severity and exploitation evidence

Critical · 10.0/10 · CVSS 3.1. Severity source.

Exploitation not established: Malicious exploitation status remains unverified in the reviewed sources.

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

Template formatting allowed traversal into private Python object attributes.

Conditions required

A caller can reach template evaluation. Default installations expose it publicly and authenticated installations allow any account.

Potential impact

Environment data and protected credentials can be exposed. A disclosed Phoenix secret can enable administrator session forgery.

Affected versions and configurations

13.0.0 through versions before 15.7.0.

Fix and mitigation status

15.7.0.

Practical next steps

Upgrade to 15.7.0 or later. Rotate exposed application secrets and provider credentials, and invalidate affected sessions.

Operating lesson

Treat template engines as a boundary around application memory.

Evidence and limits

Original disclosure 2026-11-09. The CVE Program published this record 2026-11-16. The CNA scores the issue 10.0 under CVSS 3.1. Exposure depends on the deployment’s authentication and stored secrets.

Sources