CVE-2026-2472
Vertex AI evaluation visuals could execute untrusted script in a notebook
Google Cloud · Vertex AI SDK for Python · evaluation visualization
First documented disclosure · Review checkpoint
Summary
Rendering untrusted model evaluation results could execute script in the notebook viewing them.
Severity and exploitation evidence
High · 8.6/10 · CVSS 4.0. Severity source.
Exploitation not established: Google documented a vulnerability and a correction. The reviewed sources identify no exploitation against customer environments.
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 evaluation display treated some data as active browser content.
Conditions required
A user views affected model results or dataset content in Jupyter or Colab.
Potential impact
A notebook session could execute script chosen by the supplier of the evaluation data.
Affected versions and configurations
The CNA identifies SDK 1.98.0 through versions before 1.131.0.
Fix and mitigation status
Google identifies 1.131.0 as the corrected SDK version.
Practical next steps
Update the SDK to 1.131.0 or later and treat external evaluation data as untrusted when rendering notebook output.
Operating lesson
Test display sanitization with data supplied by models and outside users.
Evidence and limits
The security fix was public December 15, 2025 and appears in release 1.131.0, published December 17 UTC. The later February 20 bulletin describes the impact and calls the release date December 16. The register uses the earlier public security-change date.
Sources
- Google Cloud security bulletin GCP-2026-011
Primary source ·
- Google Cloud CVE authority record
Primary source ·
- Google visualization vulnerability fix
Primary source ·
- Google Python SDK 1.131.0 release
Primary source ·