Machine learning security starts with the core
infrastructure, including underlying compute, storage,
and networking. When assessing infrastructure and
network security of machine learning solutions, look
for these critical qualifications: 1) the ability to isolate
the network and keep data traffic across the various
components of the workflow within secure private
network connections; 2) the ability to control access,
and, more specifically, to block inflow (ingress) and
outflow (egress) of data and code from and to the
internet; and 3) a tenancy model that provides isolation
between user environments.
Amazon SageMaker uses Amazon Virtual Private
Cloud (VPC), a service that provides logically isolated
sections of the AWS Cloud to launch its resources in
a virtual network of its own. All data traffic between
various Amazon SageMaker components flows within
this network, controlled tightly by security group
permissions. You also have the option to deploy Amazon
SageMaker within your own VPC to provide secure
access to your private resources.
In addition, Amazon SageMaker enables network
isolation from the internet by allowing you to disable
outbound data traffic to the internet through its network.
This option helps prevent users from engaging in risky
behaviors, such as installing unauthorized software.
You can also control Amazon SageMaker's network
traffic using AWS PrivateLink, a service that provides
private connectivity between VPCs, AWS services, and
on-premises applications. Further, Amazon SageMaker
instances are deployed on single-tenancy Amazon
EC2 instances to ensure that your machine learning
environments are isolated from other customers. Lastly,
Amazon SageMaker allows you to restrict root access to
users in a programmatic fashion, so you can decide when
to give your data scientists the flexibility they need to
leverage external libraries.
Infrastructure and network security
Learn more about infrastructure
security in Amazon SageMaker ›
4
