Skip to main content

Training Data or Model Artifacts are Exfiltrated

CCC.MLDE.TH02

Notebook instances and connected data pipelines may be configured with unrestricted egress paths, such as file downloads, public network interfaces, or writable external destinations. Training datasets, model artifacts, and embedded credentials could be transferred out of the environment through these paths. This results in a loss of confidentiality for proprietary data and models, and may expose regulated or sensitive records used in training.

Related Capabilities

IDTitleDescription
CCC.MLDE.CP01Managed Notebook EnvironmentsProvides fully managed notebook instances specifically designed for machine learning development, eliminating the need to manage underlying infrastructure.
CCC.MLDE.CP07Data Pipeline IntegrationSupports integration with data preparation and feature engineering pipelines, including versioning of datasets and capabilities used in ML experiments.
CCC.MLDE.CP08Model RegistryProvides centralized storage and versioning for trained models, including metadata about training runs, model artifacts, and deployment history.

Related Controls

IDTitleDescription
CCC.MLDE.CN02Disable File Downloads on MLDE InstancesPrevent unauthorized file downloads from MLDE instances to protect sensitive data from being exfiltrated.
CCC.MLDE.CN07Restrict Public IP Access on MLDE InstancesPrevent public IP access to MLDE instances to reduce exposure to the internet and enhance security.