AWS sagemaker: Warn about MLflow model sync into pre-existing Model Package Groups
Summary
Adds a warning explaining that automatic MLflow model registration syncs into a Model Package Group named from the tracking server/MLflow App ARN hash, which may pre-exist and have been created by a different IAM principal, and recommends verifying with DescribeModelPackageGroup and the CreatedBy field before enabling sync.
Security assessment
The diff documents a concrete cross-principal data-exposure weakness: deterministic naming can cause model artifacts to be synced into a package group owned/created by a different IAM principal, making the data accessible to that principal, and prescribes a CreateBy/DescribeModelPackageGroup verification as mitigation. This is an authorization/data-leak concern with a specific preventive control, hence security-related documentation.
Evidence
+A pre-existing group with the same name might have been created by a different authorized IAM principal for that account, and the data synced into that group is accessible to that IAM principal. To avoid accidentally syncing registered model data to a pre-existing group, you can verify that a SageMaker Model Package Group with such a name does not already exist by calling the [DescribeModelPackageGroup](https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeModelPackageGroup.html) API and checking the `CreatedBy` field.
Diff
diff --git a/sagemaker/latest/dg/mlflow-track-experiments-model-registration.md b/sagemaker/latest/dg/mlflow-track-experiments-model-registration.md index 757c27d..599c411 100644 --- a//sagemaker/latest/dg/mlflow-track-experiments-model-registration.md +++ b//sagemaker/latest/dg/mlflow-track-experiments-model-registration.md @@ -29,0 +30,6 @@ Model Registry sync is an opt-in feature. To enable it, call the [UpdateMlflowTr +###### Warning + +When automatic model registration is enabled, MLflow registered models are synced into a SageMaker Model Package Group with the following name: ``mlflow-registered-model-name`-`suffix``, where `suffix` is the first six characters of the SHA-256 hash of the last segment of the tracking server or MLflow App ARN (the tracking server name or the MLflow App ID). + +A pre-existing group with the same name might have been created by a different authorized IAM principal for that account, and the data synced into that group is accessible to that IAM principal. To avoid accidentally syncing registered model data to a pre-existing group, you can verify that a SageMaker Model Package Group with such a name does not already exist by calling the [DescribeModelPackageGroup](https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_DescribeModelPackageGroup.html) API and checking the `CreatedBy` field. +