AWS sagemaker high security documentation change
Summary
Added security guidance for securing serialized data storage, including explicit warnings about restricting write access to S3 locations and bucket policy recommendations.
Security assessment
Directly addresses data tampering risks by adding specific instructions to prevent unauthorized modification of serialized data through S3 bucket policies and dedicated storage locations. Evidence includes explicit warnings about access control.
Diff
diff --git a/sagemaker/latest/dg/train-remote-decorator-invocation.md b/sagemaker/latest/dg/train-remote-decorator-invocation.md index 6e1892e08..fb9858d04 100644 --- a//sagemaker/latest/dg/train-remote-decorator-invocation.md +++ b//sagemaker/latest/dg/train-remote-decorator-invocation.md @@ -342 +342,20 @@ For more information about what to include in a configuration file for a remote -Administrators can provide settings for your serialized data, including its location and any encryption settings in a configuration file. By default, the serialized data are encrypted with an AWS Key Management Service (AWS KMS) Key. Administrators can also restrict access to the root directory that you specify in your configuration file with a [bucket policy](https://docs.aws.amazon.com/AmazonS3/latest/userguide/example-bucket-policies.html). The configuration file can be shared and used across projects and jobs. For more information, see [Configuration File](https://docs.aws.amazon.com/sagemaker/latest/dg/train-remote-decorator-config.html). +By default, SageMaker AI stores serialized data in the default SageMaker AI bucket, which might be shared across projects. To prevent unauthorized modification, use the `s3_root_uri` parameter of the `@remote` decorator or `RemoteExecutor`. Specify a dedicated S3 bucket or prefix for your serialized function arguments, results, and code archives. + +###### Important + +Restrict write access to the `s3_root_uri` location to prevent tampering with serialized data. Apply a [bucket policy](https://docs.aws.amazon.com/AmazonS3/latest/userguide/example-bucket-policies.html) to your dedicated S3 location. Grant write access only to the SageMaker AI execution role used by your jobs to prevent other principals from modifying the data. + +The following example shows how to specify a dedicated S3 location with the `@remote` decorator: + + + from sagemaker.remote_function import remote + + @remote( + instance_type="ml.m5.xlarge", + s3_root_uri="s3://amzn-s3-demo-bucket/sagemaker-jobs" + ) + def my_training_function(data): + # Your training logic here + return model + +By default, AWS Key Management Service (AWS KMS) encrypts the serialized data. You can configure the storage location and encryption settings in a configuration file. You can share the configuration file across projects and jobs. For more information about the configuration file, see [Configuration file](https://docs.aws.amazon.com/sagemaker/latest/dg/train-remote-decorator-config.html).