Protect data in use with Azure Confidential Computing and Intel SGX

See how to create confidential clouds that host sensitive data in public clouds with Intel Software Guard Extension, as part of Azure confidential computing. Protect your sensitive information in the Cloud.

 

 

 

 

Mitigate privileged access attacks with hardware enforced protection of sensitive data, and protect against data exfiltration from memory. Beyond security, we’ll demonstrate machine learning analytics on multi-party data.

 

Data center security expert, Mike Ferron-Jones from Intel, gives you an exclusive look at Microsoft’s work with Intel.

 

 

 

 

Protect in-use data.

 

 

 

Keep sensitive data protected while it is actively in use in the processor and memory.

 

 

 

 

Cold boot attack by a rogue administrator? Yeah, we thought of that, too.

 

 

 

Safeguard data stored in memory with hardware-protected encryption keys.

 

 

 

 

Pool information without compromised security or privacy.

 

 

 

Work with organizations using a combined dataset — without compromising security or privacy.

 

 

 

 

Watch our video here.

 

 

 

 

 

 

QUICK LINKS:

 

 

— Introduction

 

— Protect against memory attacks

 

— Example of a cross tenant data exfiltration attack

 

— Protect your data in use: Confidential computing

 

— Mitigate privilege escalation attacks with Intel SGX

 

— New confidential computing scenarios

 

— Wrap up

 

 

Link References:

 

 

Detailed information on Azure confidential computing at Azure Confidential Computing – Protect Data In Use | Microsoft Azure

 

Watch our Zero Trust series at Zero Trust Security Implementation - Essentials Series - Episode 1

 

Keep up to date on Intel innovations at Security Innovation | Intel

 

More information on Intel SGX go to https://www.Intel.com/SGX

 

 

Unfamiliar with Microsoft Mechanics?

 

 

As Microsoft’s official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft.

 

• Subscribe to our YouTube: https://www.youtube.com/c/MicrosoftMechanicsSeries?sub_confirmation=1
  
• Talk with other IT Pros, join us on the Microsoft Tech Community: Microsoft Mechanics Blog
  
• Watch or listen from anywhere, subscribe to our podcast: Microsoft Mechanics Podcast
  
• To get the newest tech for IT in your inbox, subscribe to our newsletter: Why, How & When to use New Microsoft Tech. - Revue
  

 

Keep getting this insider knowledge, join us on social:

 

• Follow us on Twitter: https://twitter.com/MSFTMechanics
  
• Share knowledge on LinkedIn: https://www.linkedin.com/company/microsoft-mechanics/
  
• Enjoy us on Instagram: Microsoft Mechanics (@microsoftmechanics) • Instagram photos and videos
  
• Loosen up with us on TikTok: TikTok
  

Video Transcript:

 

 

- Up next, we take an exclusive look at Microsoft’s work with Intel to protect your most sensitive information in the cloud. We’ll unpack the latest silicon-level Zero Trust protections and how they help mitigate against privileged access attacks with hardware enforced protection of your most sensitive data with Intel Software Guard Extensions, plus additional defense in depth silicon-level protections against data exfiltration for memory. And beyond security, we’ll also demonstrate confidential computing scenarios that are now possible, such as machine learning analytics on multi-party data and more. And joining us to walk through all of this is data center security expert, Mike Ferron-Jones from Intel. Welcome to Microsoft Mechanics.

 

 

 

- Thanks, Jeremy, it’s a pleasure to be on the show.

 

 

 

- And it’s really great to have you on explaining another key part of the Zero Trust defense in depth story in Azure, which really spans from the silicon all the way up to the cloud.

 

 

 

- Right, and Silicon plays an integral part in a Zero Trust defense in depth strategy. At Intel, we’ve spent almost two decades creating hardware-based security innovations, and these include the protection of data held in memory as well as protections for data actively in use during the compute operations in places like the Azure cloud.

 

 

 

- And that’s really the point, because like our CTO Mark Russinovich often says, it’s your data. And as part of Zero Trust, even your cloud service provider shouldn’t be inside your own trust boundary. So for Azure’s part, we’re already providing a secure environment where we protect your data while it’s in rest in data centers, and also encrypt it while it’s in transit. And with Azure confidential computing, we take it a step further by protecting your highly sensitive data while it’s in use. And you can hold the encryption keys as well. And this is really good news, especially if you’re from a highly regulated industry or maybe you have privacy and compliance concerns over exactly where your data is stored and how it’s accessed by apps, processes, and even human operators. And these are all areas by the way that we’ve covered on Mechanics at the service level. And we have a whole series dedicated to the topic of Zero Trust at aka.ms/ZeroTrustMechanics, but as we’ll explore today, silicon-level defenses take things to the next level. So why don’t we get into this by looking really at potential attack vectors, and why don’t we start with memory attacks?

 

 

 

- Right, so a security mindset means that we’ve got to be prepared for the worst. And so the scenario I’m going to start with today may seem a bit far-fetched, but the very fact that we defend against it shows how seriously we take a defense in depth approach. Now, something a lot of people don’t know is that DRAM, even when it’s powered down, can retain its data for up to several minutes under certain environmental or temperature conditions. So imagine that a rogue admin in the data center was able to spray liquid nitrogen on the memory to freeze the DRAM, remove the DIMMs and install them into another device that dumps and stores the memories contents. Then it can be analyzed for sensitive, valuable information such as encryption keys or account credentials. This is known as a cold boot attack.

 

 

 

- And this seems pretty far-fetched, especially given all of the protections that we have for accessing Microsoft’s data centers, all the perimeter securities, etc. So it kinda seems a bit more like a mission impossible style attack. How would we stop something like this?

 

 

 

- Right, I mean, a lot of things would have to go wrong in order for this type of attack to succeed, but here’s where Silicon level protection can help prevent this attack. The latest generation of Intel Xeon Scalable Processors are equipped with a technology called Intel Total Memory Encryption, which helps protect the data stored in memory with a unique hardware protected encryption key. The Xeon memory controller encrypts the data as it’s written to the DIMM, so that even if the data is dumped, none of it is readable.

 

 

 

- And while this seems like a pretty unlikely attack, you know, it really brings home how important it is to protect data while it’s in memory. I personally had no idea that when you take the DIMMs off the board and they’re powered down, they still hold some memory that you can actually exfiltrate and get into another device. But is there something, you know, maybe doesn’t require as extreme of a situation as physical infiltration into the data center where memory protection might come into play?

 

 

 

- Sure, so let’s take an example of a cross tenant data exfiltration attack. So let’s say a sophisticated attacker poses as an Azure customer, and they set up an instance with a malicious virtual machine. Their plan is to spoof legitimate memory reads from neighboring VMs and bring the data into their malicious VM. So to succeed, they have to first get past the Azure Hypervisor, which works with the CPU’s virtualization technology to create page tables that assign separate memory regions for each VM on the DIMMs. The attacker’s VM includes an exploit that tricks the hypervisor to copy the page frame numbers from the software page table, such that the malicious VM can read or copy data from the memory regions of neighboring VMs.

 

 

 

- And to be clear, you know, an attacker would need to discover a zero-day vulnerability in the hypervisor itself. So it’s really no small feat.

 

 

 

- No, it’s no small feat indeed, but with a Zero Trust mindset, we shouldn’t underestimate the sophistication of the attacker or overestimate the invulnerability of software. So to layer on another level of protection, that’s where our Xeon processor with the Total Memory Encryption Multi-Key capability or Intel TME-MK comes into play. As the hypervisor and CPU assign memory regions to each VM, TME-MK assigns each VM its own encryption key that’s protected by the CPU’s hardware. So now if we play back the attack, even though the attacker’s VM uses the zero-day hypervisor exploit to access the memory of neighboring VMs, it can only read or copy cyphertext from memory. The stolen data is unreadable.

 

 

 

- And this will help protect against certain forms of lateral attacks like the one you just described. And I know that some Azure customers will opt to pay more for server infrastructure that’s dedicated to their organization, so by design it isn’t shared with other organizations. So using things like TME and TME-MK could be a less expensive way to get additional levels of isolation and protection. That said, though, I’d like to dig in a bit deeper, you know, into the area of partnership with Intel, you know, in terms of the protection of data while it’s in use.

 

 

 

- Right, so you’re talking about confidential computing, of course. Now this is an area of joint innovation where we focus on protecting data while it’s actively in use in the processor and memory. And Intel and Microsoft were founding members of the confidential computing consortium, which was created to move this area of computing forward through investments in technology and in building the ecosystem. To help protect sensitive data while it’s in use, Intel developed Intel Software Guard Extensions, or SGX, that create protected areas of the CPU and memory, what we call an enclave, designed to allow only verified, trusted code to process confidential data.

 

 

 

- Of course, Microsoft also built the Azure confidential computing DC series virtual machines, which leverage these silicon level innovations. So what types of attack then does Intel SGX help to mitigate?

 

 

 

- So one of the most difficult types of attack to protect against is a privileged escalation attack. Now these are most commonly software-based attacks where low-privilege code exploits vulnerabilities in high-privilege software to gain deeper access to data, to applications or the network.

 

 

 

- And equally a rogue system admin inside the organization, or a bad external actor with stolen admin creds could also have access to do reconnaissance inside the network. So how would something like Intel SGX stop here?

 

 

 

- So as we’ve touched on, Intel SGX can help mitigate these types of threats. It’s designed such that any software running outside the enclave can’t see the data and code inside. Even if it has escalated its privileges, it’s just not trusted. Now that includes any other applications, operating system, the hypervisor, even the VM and cloud administrators. In fact, Intel SGX has the smallest trust boundary of any confidential computing technology in the data center today.

 

 

 

- And that really helps mitigate against things like the rogue insider reconnaissance effort and only trusted and protected code or algorithms would be able to see and process the data. But would this work then if maybe the app was hijacked or overwritten?

 

 

 

- Mm-hmm, and this is where attestation comes in. Intel SGX aware applications have a cryptographically signed and authenticated manifest. If an attacker attempts to modify the code, the profile of the modified code won’t match the manifest provided by the original author of the software. It’ll fail attestation, which means it can’t load and it can’t access the confidential data.

 

 

 

- And you also mentioned the term SGX aware, and to that point, you know, there’s been a ton of work to make sure that apps and workloads can take advantage of Intel SGX. So Microsoft has contributed to an open enclave SDK for developers to easily build apps that can take advantage of the hardware enclave during specific operations. And we’ve also taken one of our most popular apps, SQL server, and also other derivations of SQL, and made those SGX aware as well, with something called Always Encrypted with secure enclaves, which leverages Intel SGX to run the SQL query processor inside of the enclave. And in the Azure marketplace, we’ve also published over a dozen different solutions provided by ISVs. That said, though, why don’t we look beyond the different attack mitigations? Why don’t we switch gears to something that might light up as part of using confidential computing scenarios?

 

 

 

- All right, well, that’s the really exciting part. This opens up new ways for different organizations to work together on shared datasets in multi-tenant public cloud services without compromising security or privacy. I’ll show you an example here where two banks want to combine their individual datasets to perform a fraud analysis on a larger pool dataset. Now by combining their data, they can increase the precision of the fraud detection machine learning model, so that both banks benefit without exposing their transaction data to the other bank or to the cloud operators. Now here you can see, I have three windows open. On the left side, I’m logged into two different bank environments. The one on the right is from a virtual machine in Azure. Again, each bank environment has its own private dataset that it owns and controls. Now I’ll start with a baseline. I’m going to kick off the fraud analytics inference detection on top of bank one’s dataset. And I get a precision of around 92.7%, and each bank will get a similar result on their individual dataset. Now, the challenge is that this is regulated data and that each bank would want to protect it from access by the cloud provider, as well as the other banks.

 

 

 

- How much better does it get then when bank one and bank two combine their datasets?

 

 

 

- And that’s what we’re trying to solve for with the right security guardrails in place, of course. Now I’ll combine the datasets from both banks and perform the same fraud analytics from before, using the VM running in Azure. Now, once it completes, you’ll see that the precision increases to 98.2% versus the 92.7% that we got just using bank one’s data alone. And this sounds promising, but I perform this computation using data from two unencrypted data files.

 

 

 

- And right now the data sharing model between the banks and the operator isn’t ideal. So how can we add more protection to that?

 

 

 

- Yes, so since the data files weren’t encrypted, each bank’s data could be visible to the other bank. It could also be visible to an intruder in their shared VM that hosts the fraud detection model or the VM’s memory. And from a confidentiality and regulatory perspective, this just isn’t going to cut it. In this cloud operator window, I’ll first look at the processes run in the Azure VM. And then I identify the process I just ran, which is 17,838. I can dump its memory contents, and we can see that the data that we want to protect is in the clear and vulnerable to anyone with access to what ran in memory, whether that’s a rogue insider, or an attacker who manages to breach the infrastructure.

 

 

 

- So how can we make this even more secure?

 

 

 

- Well, let’s run that same computation using Intel SGX enclave. So in this case, I’ll use encrypted data files containing the same data that we just used from bank one and bank two. Now I’ll launch the app using Intel SGX and an open-source library OS called Gramine that allows an unmodified app to run in an SGX enclave. In doing this, only the SGX enclave has access to the encryption keys needed to process the data from the encrypted CSV files. Now, first I’ll run the analytics one more time. And as you’d expect, we get the same precision as before, 98.2%. But this time the memory is protected. If I go back to the view from the operator window, and I look at the process list again, and for that last run, it was 17957. Now I’ll dump the memory from this process and you’ll see that there’s not any visible sensitive data in the memory dump at all. Only the protected processes inside the enclave can see the data and process it.

 

 

 

- And importantly, in this case, both banks actually have the advantage of higher precision analytics. The data was protected all the time end-to-end, including while it was running in memory.

 

 

 

- Right, and this is a big advantage for both banks, because it’s really hard to do fraud detection on your own, especially when the potential violators are hopping from bank to bank to bank. And this is just the tip of the iceberg. There are so many more confidential computing scenarios across a range of industries. You know, these often involve multi-party computing on shared or regulated data. Now this could be everything from disease diagnostics in healthcare involving multiple hospitals, high security information sharing within or across governments, or to secure payment processing, including credit card or bank transactions, just to name a few.

 

 

 

- And Intel SGX, along with Azure confidential computing, makes it a lot easier to create confidential clouds inside the public cloud to host your most sensitive data. So for anyone who’s watching, looking to build solutions on Intel SGX, or learn more about memory encryption technologies, what do you recommend?

 

 

 

- Right, so to learn more and keep up-to-date with Intel innovations in this space, you can go to intel.com/security and to learn more information about Intel SGX, go to Intel.com/SGX.

 

 

 

- Thanks, Mike, and also be sure to check out aka.ms/AzureCC for more information on Azure confidential computing. Of course, keep watching Microsoft Mechanics for all latest tech updates. Subscribe to our channel if you haven’t already, and as always, thank you for watching.

 

Continue reading...