Disclosed weakness · no CVE recorded
GPUStack could execute repository code while inspecting a model
GPUStack · GPUStack
First documented disclosure · Review checkpoint
Summary
Submitting a model for inspection could cause GPUStack to run code from the model repository with the server’s permissions.
Severity and exploitation evidence
Critical · 9.1/10 · CVSS 3.1. Severity source.
Exploitation not established: The original source establishes the weakness but does not report malicious exploitation.
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
Model metadata inspection could run repository-supplied code with the inference management server’s privileges.
Conditions required
A user can submit a model for evaluation or deployment and influence its repository content. Required platform roles vary by version.
Potential impact
Data and credentials available to the GPUStack server could be exposed; service state or availability could change.
Affected versions and configurations
>= 2.0.0, < 2.3.1
Fix and mitigation status
2.3.1 fixes this issue.
Practical next steps
Upgrade to 2.3.1 or a later release containing this fix. Restrict model submission to trusted operators and review model sources. Use the patched server even for metadata inspection.
Operating lesson
Inspect model configuration as data before granting any model code permission to run.
Evidence and limits
The maintainer assigns the score. No CVE was listed in the original advisory at review.
Sources
- GPUStack security advisory GHSA-gjf9-3p3r-9h9c
Primary source ·
- GPUStack 2.3.1 release notes
Primary source ·