What Is Confidential Compute and Why You Want It
What Is Confidential Compute and Why You Want It
How Encryption of ‘Data in Use’ Changes the Game
Data encryption is an essential ingredient of information security. We all know the ACIA1 concept:
Authorization: stopping entities from going into places
Confidentiality: stopping entities from looking at things
Integrity: stopping entities from moving and altering things
Availability: stopping entities from interrupting processes
Encryption is a technique used to achieve Confidentiality, i.e. stopping entities from looking at data for which they have no authorization.
In digital security, data exists in 3 modes: in motion, at rest, and in use.
Data in motion
Imagine a laptop, smartphone, or another computer system running an application. That app connects to a backend (aka server). The connection between the app and the server is used to transfer data. The data travels over this digital highway.
This is data in motion. And you want it encrypted so that no 3rd parties (the client and the server are 1st and 2nd parties) can eavesdrop. Think of https, which uses TLS/SSL to encrypt data in motion.
In reality, we’ve used the internet for 15 years, 1990-2005, without any encryption of data in motion. We just sent the naked data over the wire, implicitly trusting all intermediate parties not to read our data.
Today, you wouldn’t use a website without https.
Data at rest
The second mode is data at rest. This data is stored somewhere—usually a database, a disk, blob storage, or file share.
For a few years, public cloud providers have encryption of ‘data at rest’ by default, which is good. They use platform-generated encryption keys so, in theory, they can access your data. You can choose to bring your own encryption keys. The most used encryption algorithm is AES-256.
Data in use
The third mode is data in use. Also known as data in memory. This is when the backend application processes your data. This computing or processing happens in memory while the app is running.
Here’s the killer: data in use is NOT encrypted.
This means all data being processed in memory is unencrypted and plainly available. This means that private and public providers can read your data in memory.
So you might think you have strong encryption for your transport and storage layers, while in reality, your data is exposed and vulnerable for the duration it resides in memory. This poses a severe information security risk.
The missing puzzle piece
With encryption of data in motion and at rest already encrypted, encryption of data in use is the missing puzzle piece for security architects. I call it 2-point vs 3-point encryption.
Imagine you want your house protected, so you reinforce the front and back doors and apply special burglar-proof glass in your windows. But you leave a window on the ground floor unlocked. It’s not directly visible, and you need to look, but it’s open. Yes, you applied protection, but the weakest link is your maximum security.
Now, imagine processing sensitive data on a private or public cloud. How do you know your data is not secretly copied from memory?
Enter Confidential Computing
Confidential computing is a technique that enables data encryption in memory. Your data remains encrypted even while it’s being used. It’s 3-point encryption, and it’s the missing puzzle piece.
How does this magic work, you ask? It requires special hardware. We call it encryption on the silicon level. The three most prominent vendors of confidential compute chips are:
Intel (CPU)
AMD (CPU)
NVIDIA (GPU - for confidential AI applications)
To be clear, this type of encryption is on the computing layer, NOT storage or networking. Those are already covered.
Just like https is now the standard, and we can’t think of using websites without it, confidential computing will become the norm. I expect in less than five years.
There will come a day when we’ll say: remember when we processed sensitive data in memory unencrypted?
The Catch
At the time of writing, I see the following catches with confidential computing:
The confidential computing hardware is almost unavailable for regular organizations because the hyperscalers (AWS, Azure, GCP) are buying up all the hardware to make it available through their Public Cloud. So, using it means moving to a Public Cloud. However, I predict the availability of confidential compute-powered devices in the next 24-36 months.
Although improving every 12 months, some current iterations of confidential computing require changes to your apps to run on this special hardware (Intel SGX, AMD SEV). But new iterations are on the horizon that require fewer to no app changes (Intel TDX).
There’s a performance penalty (10-30%) when using confidential computing compared to regular computing. You can compensate for this by choosing a more performant Virtual Machine.
So what?
As a CIO or CISO, you want to start investigating confidential computing to host your sensitive workloads that process sensitive data. As confidential computing is becoming more mainstream, you have a new defensive weapon in your arsenal to increase your security posture.
Jetro WILS is the founder of BlueDragon Security, where he helps organizations operate safely in this cloud era by strengthening their digital security and compliance.
Disclaimer: This article is written in a personal capacity, and the views expressed are solely mine. They do not represent the positions, strategies, or opinions of my clients, who bear no responsibility for this content.
The CIA-triad is a subset of the ACIA-quad.