Skip to main content

AWS Sagemaker Security Patterns- Part 1

We are going to look into some best security architecture of AWS Sagemaker. By default AWS Sagemker is a managed sevice and everything is managed my the AWS, and not the customer.

Image courtesy- AWS

Secure Deployment patterns:
1. Default architecture: The managed sagemaker VPC is directly connected to the Internet, which is not an ideal architecture from security point of view.

2. The customer can disable the direct Internet connection route it back through their own VPC. Here all the traffic is not being routed through a VPC that's controlled by the customer. The notebook cannot be accessed via open Internet now:

 3. Restrict the access to the notebook IAM policy based on IP/ VPC, so that no one can directly access the Notebook via Internet such as through their personal laptop.
The resultant architecture of above restriction. The user is forced to access the Notebook via corporate IP ranges:

4. Sagemaker default training architecture. This also run in the managed account in managed VPC, where everything goes through Internet gateway, which is not desirable from security prespective.

5. And this is how a more secure architecture looks like, where the Internet connection is turned off and the traffic goes via the customer's traffic. This is achieved by creating an ENI back in customer's account. The cluster of training instances deployed create one ENI ecah for all the instances to the customer's VPC and that's how the traffic is routed to the customer's VPC:
6. Also while doing training job, the VPC configurations must be set. This can be automated as well.

7. Hosting default architecture- all traffic going via Internet:

8. Recommended secure hosting deployment:


9. Recommended restrictive VPC endpoint IAM policy:


Comments

Popular posts from this blog

Using an AirPcap device in Windows with Wireshark

Capturing wireless traffic in a Windows environment is unfortunately not as easy as a setting change. As with most Windows-based software, drivers in Windows are often not open source and do not allow for configuration change into monitor mode. With this in mind, we must use a specialized piece of hardware known as an AirPcap device. Once you have obtained an AirPcap device you will be required to install the software on the accompanying CD to your analysis computer. The configurable options include: • Interface - Select the device you are using for your capture here. Some advanced analysis scenarios may require you to use more than one AirPcap device to sniff simultaneously on multiple channels. • Blink LED - Clicking this button will make the LED lights on the AirPcap device blink. This is primarily used to identify the specific adapter you are using if you are using multiple AirPcap devices. • Channel - In this field, you select the channel you want AirPcap to listen on. Extension C...

Anti CSRF header

Recently I came across an application which was preventing crsf attacks using a unique non-traditional approach. In traditional approach the csrf is thwarted by embedding unique random tokens, called nonce, in each sensitive page. But this application, which was making ajax calls and used jQuery, was creating a header to identify the valid and invalid requests altogether. The idea is to generate a custom header, x-session-token in this case, with every request which is considered sensitive and includes any sort of transaction. For example: xhr.setRequestHeader('x-session-token', csrf_token)   At the server level, server checks for this header if found request is fulfilled, otherwise rejected. We need to use xhr calls for making use of this technique, not useful in regular POST and GET requests. Since, I was not aware of this kind of countermeasures, probably, since most of the applications I did were using standard requests. So, I searched a bit and found even Go...

Some one watching where you visited!

Yes... Mozilla has been susceptible to browser-history stealing java script code. Today, Giorgio posted some cool information about the exploit. Mozilla is already working on this. This bug has been reported. Actually they have set up a web site to show the proof-of-concept. Visit www.statrpanic.com in FF,Safari or Netscape and it will tell you which websites have you been already ! But I am not sure it will work in IE or not because my IE is not responding to the website. Clearing history of visited website makes you safe to this attack. I mean this is one way..may be there are other ways to exploit this. But I have found this effective. Try it yourself in FF and then in IE and see the results.