Accelerating machine

learning innovation
through security
Security features from Amazon SageMaker and the
AWS Cloud can help you go from idea to production faster.
I N T R O D U C T IO N

How security helps deliver machine

learning results
To build successful machine learning models, you often need datasets unique
to your business. These datasets are extremely valuable assets and need to
be secured throughout every step of the machine learning process—including
data preparation, training, validation, and inference.

In a typical machine learning project, it can take months to build a secure

workflow before you can begin any work on your models. Maintaining
executive buy-in means delivering fast results— so accelerating projects by
weaving security into every step of the process will help ensure organization-
wide commitment to your project and your larger machine learning initiatives.

Amazon SageMaker is a fully managed service that provides every developer

and data scientist with the ability to build, train, and deploy machine learning
models quickly and securely. In this eBook, we provide an overview of the
Amazon SageMaker security features that can help your organization meet
the strict security requirements of machine learning workloads—ultimately
helping you go from idea to production faster, more securely, and with a
higher rate of success.

2
Executive summary
As a managed AWS service, Amazon SageMaker automatically inherits the AWS global infrastructure and its network security features. AWS is
purpose built for the cloud, with data centers and a network architected to help protect your information, identities, applications, and devices.
The AWS network and infrastructures are monitored 24/7 to ensure confidentiality, integrity, and availability of your data. In addition, Amazon
SageMaker offers a comprehensive set of capabilities, so you can run your machine learning workloads with the most flexible and secure machine
learning environment available today.

Customers have told us that the following are the key security criteria they consider when evaluating machine learning solutions. Together, AWS
Cloud and Amazon SageMaker security features allow you to meet these criteria readily—so you can put machine learning to work securely in
production applications.

Infrastructure and Data protection Compliance Authentication and Monitoring and

network security certifications authorization auditability
Get automatic data
Control data traffic across encryption at rest and in Inherit the most Define, enforce, and audit Track, trace, and audit
Amazon SageMaker transit with flexibility to comprehensive compliance who can be authenticated all API calls, events, data
components over a bring your own keys. controls and easily meet and authorized to use access, and interactions
private network. Ensure your industry’s regulatory Amazon SageMaker down to the user and
appropriate ingress/egress requirements. resources. IP levels.
with single-tenancy, so
your data and resources
are secure.

3
Infrastructure and network security
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.

Learn more about infrastructure security in Amazon SageMaker ›

4
3M innovates while maintaining focus

Through research and development, 3M introduces more than 1,200 new

products every year. The company is using Amazon SageMaker to improve the
effectiveness of its quality control processes and move beyond time-consuming
manual inspection. 3M built machine learning models to improve materials
research, analysis, and defect detection.

“Our new machine learning-based processes are far

more efficient than prior approaches. (Amazon)
SageMaker provided cost-effective access to powerful
infrastructure on demand, along with comprehensive
security features. This allowed us to focus on research
rather than the mechanics of securely scaling our
compute capabilities.”

David Frazee, Vice President, 3M Corporate Research Systems Lab

5
Authentication and authorization
One of the fundamental capabilities you need to secure your machine learning environment is a strong mechanism to define, enforce, and audit
who can sign in (called authentication) and what resources and functions they are authorized to access (called authorization).

Amazon SageMaker is governed by AWS Identity and Access Management (IAM), a service that enables you to manage access to AWS services and
resources securely. With AWS IAM, you can implement fine-grained access controls. AWS IAM allows you to specify who can perform what actions to
which resources and under what circumstances at the level of specific features, users, groups, and roles. You can readily bring existing user identities
from AWS Directory Service, your enterprise user directory such as Active Directory (AD), Lightweight Directory Access Protocol (LDAP), or a web
identity provider.

1 2 3 4
Multi-factor authentication (MFA), Tag-based access control to Detective controls that identify Preventive controls that can stop a
which prompts users for their categorize resources by purpose, potential security threats or potentially harmful action before
user name and password (the first owner, environment, and other incidents using user behavior in it takes place
factor), and an authentication code criteria, making it easier to manage, Amazon SageMaker
from their AWS MFA device (the search, and filter resources
second factor)

Learn more about AWS IAM for Amazon SageMaker ›

6
Data protection
Another important security requirement for machine learning solutions is protecting data
through automatic encryption at rest, in transit, and during training across distributed
clusters. Machine learning solutions should also provide the flexibility to bring your own
encryption keys.

Amazon SageMaker comes with built-in encryption capabilities to ensure that training
datasets, input data for inference, and other machine learning model and system artifacts
are encrypted in transit and at rest. Amazon SageMaker also gives you flexible data
encryption options through Amazon SageMaker managed keys, AWS managed keys,
and customer managed keys.

Learn more about data protection in Amazon SageMaker ›

7
The NFL tackles player safety

Together, the NFL and AWS are leveraging machine learning to build the
“Digital Athlete,” a platform to improve injury prevention and treatment—and,
ultimately, predict injury. The program will use anonymized and aggregated
player data to create a composite that will simulate infinite scenarios of the
game environment. The NFL and AWS hope the program will eventually have
implications beyond football—for example, it could become a useful tool in
the healthcare industry.

“Since the data used in the modeling is highly sensitive,

we needed an ML solution like Amazon SageMaker
with security and compliance features built-in to
protect the data throughout the ML process.”

Jennifer Langton, SVP Health and Safety, NFL

8
Monitoring and auditability
Auditability is about tracking, tracing, and monitoring API calls, events, data access, and interactions down to the user and IP levels to ensure quick
remediation (if necessary). It’s critical to be able to capture audit trails at the granular level of users, files, and objects.

Amazon SageMaker is integrated with Amazon CloudWatch Logs and AWS CloudTrail for logging events and API calls. You can also set alarms that
watch for certain thresholds and send notifications or take actions when those thresholds are met. And you can identify which users and accounts
called AWS, the source IP address from which the calls were made, and when the calls occurred. Since Amazon SageMaker uses data from Amazon
Simple Storage Service (Amazon S3), all data access activities are automatically logged for monitoring.

Learn more about logging and monitoring in Amazon SageMaker ›

“We chose (Amazon) CloudWatch because it has simplified the

process to aggregate various types of log streams from various AWS
services, and reduced the overhead costs and complexity generally
associated with logs management. (Amazon) CloudWatch also supports the
ability to export certain logs via log streams—which enabled us to set up
an external automated log analysis pipeline—so it was a clear winner.”

Moe Abbas, Cloud Team Lead, Canva

9
Regulatory compliance
In many cases, machine learning solutions need to comply with regulatory standards and
pass compliance certifications that vary significantly across countries and industries.

AWS supports more security standards and compliance certifications than any
other cloud vendor. As an AWS service, Amazon SageMaker complies with a wide
range of compliance programs, including PCI, HIPAA, SOC 1/2/3, FedRAMP, and ISO
9001/27001/27017/27018. In addition, to aid your compliance efforts, AWS regularly
achieves third-party validation for thousands of global compliance requirements across
finance, retail, healthcare, government, and more. For the latest SageMaker certifications,
see the AWS Compliance Program website.

10
Thomson Reuters innovates faster
with security

Thomson Reuters is a leading source of intelligent and trusted information

for businesses and professionals. Using Amazon SageMaker, Thomson Reuters
accelerated the development of machine learning models for a number of
innovative solutions—including text classification and natural language
question answering—with cost savings and flexibility. To complement Amazon
SageMaker’s security features, the company built a custom solution it calls
Secure Content Workspaces (SCW).

“Amazon SageMaker saved our team countless hours

of coding that would have been necessary on self-
managed ML infrastructure. Together, Amazon
SageMaker and SCW make it possible for research
and data scientists to work in the cloud in compliance
with our standards without being cloud experts.”

John Duprey, Senior Director of Engineering, Center for AI and Cognitive

Computing, Thomson Reuters

11
Try Amazon SageMaker
for two months, free
Amazon SageMaker can help your organization secure your machine
learning environment quickly—so you can focus on scaling and innovating
faster. As part of the AWS Free Tier, you can get started with Amazon
SageMaker for free.

Start your free trial ›

12