AWS
Google CloudOverview
Model artifacts can contain proprietary model weights, embedded data patterns, configuration, or sensitive training outputs. Production AI services should load model artifacts from private, access-controlled storage instead of generic public HTTP locations.
Remediation guidance
Remediation
Move model artifacts to private cloud storage with least-privilege access, encryption, and audit logging. Avoid direct public HTTP artifact URLs for production models.
- Copy artifacts into an approved private bucket or artifact registry.
- Restrict access to the model deployment identity.
- Enable encryption, logging, and lifecycle policies for model artifact storage.
Query logic
These are the stored checks tied to this control.
AI model artifacts should use private storage
Connectors
AWSGoogle Cloud
Covered asset types
AI ServicesModelStorage
Expected check: eq []
{
sageMakerModels(where: { primaryContainerModelDataURL_MATCHES: "(?i)^https?://.*" }) { ...AssetFragment }
vertexAIModels(where: { artifactURI_MATCHES: "(?i)^https?://.*" }) { ...AssetFragment }
}
